DISPATCH Working Group J.J. Garcia Aranda J. Perez Lajo L.M. Diaz Vizcaino Internet Draft Alcatel-Lucent Intended status: Standards Track November 8, 2010 Expires: May 2011 The Quality Hypertext Transfer Protocol draft-aranda-dispatch-qhttp-00.txt Status of this Memo This Internet-Draft is submitted to IETF in full conformance with the provisions of BCP 78 and BCP 79. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet- Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html This Internet-Draft will expire on April 8, 2010. Copyright Notice Copyright (c) 2010 IETF Trust and the persons identified as the document authors. All rights reserved. This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Garcia Aranda Expires May 8, 2011 [Page 1] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 Abstract This memo describes an application level protocol for the standard communication of e2e QoS compliance information using a protocol based on Hypertext Transfer Protocol (HTTP), which forms the basis for the World Wide Web, and Session Description Protocol (SDP). Quality HTTP (Q-HTTP) provides a mechanism for latency, jitter, bandwidth an packet loss negotiation and monitoring, alerting whenever one of the negotiated conditions is violated. Implementation details on the actions to be triggered upon reception/detection of QoS alerts exchanged by the protocol are out of scope of this draft, it is application dependant (e.g. increase quality, reduce bit-rate) or even network dependant (e.g. change connection's quality profile). Table of Contents 1. Introduction................................................4 1.1. Motivation.............................................5 1.2. Summary of Features.....................................6 2. Terminology.................................................7 3. Overview of Operation........................................7 3.1. Protocol Phases.........................................7 3.1.1. Handshake Phase....................................8 3.1.1.1. Description of Quality parameters inside SDP..11 3.1.2. Quality negotiation phase.........................15 3.1.2.1. Measurement of latencies and jitters..........16 3.1.2.1.1. constraints not reached.................21 3.1.2.1.2. Constraints not reached with Policy server involved..........................................24 3.1.2.1.3. Constraints reached.....................25 3.1.2.2. Measurement of bandwidth and packet loss......28 3.1.2.2.1. constraints not reached.................31 3.1.2.2.2. Constraints not reached with Policy server involved..........................................35 3.1.2.2.3. Constraints reached.....................35 3.1.2.3. Qos Level out of range.......................36 3.1.2.4. Qos Level increments without changes in network behaviour............................................38 3.1.2.5. Trigger an application in combination with HTTP38 3.1.3. Continuity phase..................................39 3.1.3.1. Normal mode..................................40 3.1.3.2. Sliding window mode..........................42 Garcia Aranda Expires May 8, 2011 [Page 2] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 3.2. Dynamic constraints and flows..........................44 3.3. QoS-level downgrade operation..........................45 3.4. Sanity check of Quality sessions.......................46 4. Q-HTTP messages............................................47 4.1. Requests..............................................47 4.2. Responses.............................................48 4.3. Header Fields.........................................50 4.3.1. Specific Q-HTTP Request Header Fields.............50 4.3.2. Specific Q-HTTP Response Header Fields............51 4.4. Bodies................................................51 4.4.1. Encoding.........................................52 5. General User Agent behavior.................................52 5.1. Roles.................................................52 5.2. Multiple Quality sessions in parallel..................53 5.3. General client behavior................................54 5.3.1. Generating requests...............................55 5.4. General server behavior................................55 6. Q-HTTP method definitions...................................56 6.1. BEGIN.................................................57 6.2. GET...................................................57 6.3. READY.................................................57 6.4. PING..................................................58 6.5. DATA..................................................58 6.6. QOS-ALERT.............................................58 6.7. CANCEL................................................59 7. Response codes.............................................59 7.1. 100 trying............................................59 7.2. 200 OK................................................59 7.3. Redirection 3xx........................................60 7.4. Request Failure 4xx....................................60 7.4.1. 400 Bad Request...................................60 7.4.2. 404 Not Found.....................................60 7.4.3. 405 Method Not Allowed............................60 7.4.4. 406 Not Acceptable................................60 7.4.5. 408 Request Timeout...............................60 7.4.6. 412 A precondition has not been met...............61 7.4.7. 413 Request Entity Too Large......................61 7.4.8. 414 Request-URI Too Long..........................61 7.4.9. 415 Unsupported Media Type........................61 7.4.10. 416 Unsupported URI Scheme.......................61 7.5. Server Failure 5xx.....................................61 7.5.1. 500 Server Internal Error.........................61 7.5.2. 501 Not Implemented...............................61 7.5.3. 503 Service Unavailable...........................62 7.5.4. 504 Server Time-out...............................62 7.5.5. 505 Version Not Supported.........................62 7.5.6. 513 Message Too Large.............................62 Garcia Aranda Expires May 8, 2011 [Page 3] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 7.6. Global Failures 6xx....................................62 7.6.1. 600 session not exist.............................62 7.6.2. 601 quality level not allowed.....................63 7.6.3. 603 Session not allowed...........................63 7.6.4. 604 authorization not allowed.....................63 8. Implementation Recommendations..............................63 8.1. Default client constraints.............................63 8.2. Bandwidth measurements.................................63 8.3. Packet loss measurement resolution.....................64 8.4. qos-level dictionary...................................64 8.5. Measurements and reactions.............................64 8.6. Scenarios.............................................64 8.6.1. Client to ACP.....................................65 8.6.2. Client to client..................................65 9. Security Considerations.....................................66 10. IANA Considerations........................................66 11. Conclusions...............................................69 12. References................................................70 12.1. Normative References..................................70 12.2. Informative References................................71 13. Acknowledgments...........................................72 14. Authors' Addresses........................................73 1. Introduction The World Wide Web (WWW) is a distributed hypermedia system which has gained widespread acceptance among Internet users. Although WWW browsers support other, preexisting Internet application protocols, the native and primary protocol used between WWW clients and servers is the HyperText Transfer Protocol (HTTP) (RFC 2616 [1]). The ease of use of the Web has prompted its widespread employment as a client/server architecture for many applications. Many of such applications require the client and the server to be able to communicate each other and exchange information with certain quality constraints. Quality in communications at application level consists of four measurable parameters: o Latency: The time a message takes to travel from source to destination. It may be approximated to RTT/2 (Round trip time), assuming the networks are symmetrical. o Jitter: latency variation. There are some formulas to calculate Jitter, and in this context we will consider the statistical variance formula. Garcia Aranda Expires May 8, 2011 [Page 4] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 o Bandwidth: To assure the quality, a protocol MUST assure the availability of bandwidth needed by the application. o Packet loss: The percentage of packet loss is closely related to bandwidth and jitter. Affects bandwidth because a high packet loss implies sometimes retransmissions that also consumes extra bandwidth, other times the retransmissions are not achieved ( for example in video streaming over UDP) and the information received is less than the required bandwidth. In terms of jitter, a packet loss sometimes is seen by the destination like a larger time between arrivals, causing a jitter growth. Q-HTTP provides a mechanism for quality monitoring and it is based on HTTP and SDP in order to be easily integrated in WWW, but it may be used by any type of application, not only those based on HTTP. Quality requirements may be needed by any type of application that communicates using any kind of protocol, especially those which have real-time constraints. Q-HTTP is an application level Client/Server protocol which pretends to measure continuously session quality for a given flow (or set of flows), end-to-end and in real-time; raising an alert if quality parameters are below a given threshold. The thresholds of each application are different, depending on the nature of each application. Q-HTTP does not describe either the actions carried out to deal with the alert or how to implement them. Q-HTTP is session-independent from the application flow/s, in order to not impact them. To perform the measurements, two control flows are created in each direction (forward and reverse). 1.1. Motivation Monitoring quality of service (QoS) in computer networks is useful for several reasons: o Enable real-time services and applications to verify whether network resources achieve a certain QoS level. o Monitoring helps real-time services and applications to run on the cloud, allowing the existence of Application Content providers (ACPs) which offer guaranteed real-time services to the final users. o Monitoring also applies to Peer to Peer (P2P) real-time applications Garcia Aranda Expires May 8, 2011 [Page 5] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 o Enable ISPs to offer QoS to any ACP or final user application in an accountable way o Enable e2e negotiation of QoS parameters, from any ISP to any ISP. A protocol to monitor QoS must address the following issues: o Must be ready to be used by current standard protocols and applications, without forcing a change on them. o Must have a formal and compact way to specify quality constraints of the desired application to run. o Must have measurement mechanisms avoiding application disruption. o Must have specific messages to alert about the violation of quality constraints in different directions (forward and reverse), because network routing may not be symmetrical, and of course, quality constraints may not be symmetrical. o Must Protect the data (constrains, measurements, QoS levels asked to the network) in order to avoid malicious measurements. 1.2. Summary of Features Quality HTTP is a message-oriented communication protocol designed to be used in a similar way like HTTP (RFC 2616 [1]). Q-HTTP can be used in conjunction with HTTP too since it is designed to coexist with HTTP's messaging model and to be easily integrated with HTTP applications. It is based on HTTP and SDP (RFC 4566 [2]) for easy integration in WWW. The benefits in quality provided by Q-HTTP can be used by any type of application which uses any type of protocol for data transport. Quality HTTP provides a quality monitoring mechanism to any communication that takes place between the client and the server, not only the Q-HTTP communication itself. Q-HTTP does not establish multimedia sessions and it does not transport application data. The type of use and kind of protocol of this quality communication is application dependant and can be whatever. Q-HTTP doesn't force any particular protocol or way of using of the quality connection. Garcia Aranda Expires May 8, 2011 [Page 6] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 Q-http defines three phases with different purposes, and inside these phases a negotiated measurement procedure is used. Different measurement procedures can be used inside Q-HTTP (although for compatibility reasons a default measurement mechanism is defined). Basically, Q-HTTP only defines how to transport SLA information and measurement results as well as providing some mechanisms for alerting. Q-HTTP MUST be executed just before starting a client-server application which needs a quality connection in terms of latency, jitter, bandwidth and packet loss. Once client and server have succeeded in establishing communication under quality constraints, the application can start, and Q-HTTP continues measuring and alerting. During the lifetime of the quality session, the protocol keeps in a special state in which it periodically renews the session and alerts if the measurements of quality parameters does not meet the negotiated application requirements. The quality parameters can be suggested by the client in the first message, but the server can accept these parameter values or force others. The server is in charge of deciding the final values of quality connection. 2. Terminology The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 [3]. 3. Overview of Operation This section introduces the basic operations of Q-HTTP using simple examples. This section is tutorial in nature and does not contain any normative statements. 3.1. Protocol Phases All elements of the IP network contribute to the quality in terms of latency, jitter, bandwidth and packet loss. All this elements have their own quality policies in terms of priorities, traffic mode, etc. and each element has its own way to manage the quality. The purpose of a quality connection is to establish an end-to-end communication with enough quality for the server application. Garcia Aranda Expires May 8, 2011 [Page 7] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 To monitor negotiated SLA compliance, three phases are defined o The handshake in which the server is contacted by the client and in the answer message it communicates the quality constraints for a given application. o The negotiation phase, in which the quality of the connection is measured in both directions (latency, jitter, bandwidth and packet loss), and Q-HTTP messages are sent in order to alert when the quality does not match the constraints. This phase is iterative until quality constraints are reached or the session is cancelled after checking that the quality constraints are impossible to reach. Just after reaching the quality requirements, Q-HTTP provides a simple mechanism to trigger optionally the application using HTTP. o The continuity phase, in which periodically, the quality is measured. If the quality measurement results become degraded, a new negotiation phase is started. In this phase the measurements MUST avoid disturbing application by consuming network resources. +------------------------------------------------+ | | | Handshake ---> Negotiation +--> Continuity--+ | | A | | A | | | | | | | | | | +--+ | +----------+ | | | | | +->Application | | starts... | | | +------------------------------------------------+ Figure 1 Phases. 3.1.1. Handshake Phase The first phase consists of a Q-HTTP BEGIN message sent from the client to the server. This message goes through all elements belonging to one or more IP networks. The first Q-HTTP message MUST have a special URI (RFC 3986 [4]), which forces the use of Q-HTTP protocol if it is implemented in a general web browser. The http URI scheme MUST be: Garcia Aranda Expires May 8, 2011 [Page 8] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 "httpq:" "//" host [":" port] [path["?" query] instead of the conventional http schema: "http:" "//" host [":" port] [path["?" query]] Optionally, the client can send the desired quality parameters (to do that, enclosed in the body of the message a SDP should be sent) and the server can take them into account when it builds the answer with the final values, following a offer / answer schema (RFC 3464 [5]). The description of these quality parameters is encoded in SDP. The server MUST answer with a Q-HTTP 200 OK message, and in the body of the answer message, a SDP MUST be included, with information of the required quality constraints. Q-HTTP responses should use the protocol designator "Q-HTTP/1.0". After these two messages are sent, the first phase is completed. The quality parameters have been sent to the client. Next step is to measure the quality of the communication path between client and server and alert if SLA is being violated. +------------------------------------------------+ | | | Client Server | | | | ------- Q-HTTP BEGIN ------------> | | | | <------ Q-HTTP 200 OK ------------ | | | | | +------------------------------------------------+ Figure 2 handshake. Example of Client Request and server answer: Client Request: ========================= BEGIN httpq://www.example.com Q-HTTP/1.0 Content-Type: application/sdp User-Agent: qhttp-ua-experimental-1.0 Content-Length: 142 (SDP not shown) ========================= Garcia Aranda Expires May 8, 2011 [Page 9] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 Server Answer: ========================= Q-HTTP/1.0 200 OK Date: Mon, 10 Jun 2010 10:00:01 GMT Content-Type: application/sdp Expires: 3000 Q-HTTP-Resource:httpq://www.example.com/example/util/agent?num=666 Q-HTTP-policy-server:httpq://www.qosmanager.com/agent Signature: 6ec1ba40e2adf2d783de530ae254acd4f3477ac4 Content-Length: 131 (SDP not shown) ========================= The header "Expires" purpose is to provide a sanity check and enables the server to close inactive sessions. If the client does not send a new request before the expiration time, the server can close the session. The header "Signature" contains a digital signature that can be used by the network to validate the SDP, preventing security attacks. The signature is an optional header generated by the server using a hash and encryption method such as MD5 (RFC 1321 [6]) and RSA (RFC 2437 [7]), but it depends on the certificate used by the server. This certificate is supposed to be delivered by a Certification Authority (CA) or policy owner to the server. The signature is applied to the SDP body. Signature= RSA ( MD5 (),) If the signature is not present, other validation mechanism may be implemented in order to provide assured quality with security and control. The optional response header "Q-HTTP-Resource-Server" contains the URI in charge of this session. This URI MUST be invoked by the client in all later requests. Example: Q-HTTP-Resource- server:httpq://www.example.com/example/util/agent?num=666 If this header is not present, the client will continue sending all requests to the original invoked URI, but if it is present, its use is mandatory. Garcia Aranda Expires May 8, 2011 [Page 10] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 The last optional response header is "Q-HTTP-policy-server" which contains the URI towards which client MUST send the later QOS-ALERT messages. This header will be explained later on. In case this header is present, the Q-HTTP-Resource-server header is mandatory. During the next phases of the protocol, the client role is not pure, but a mix of client and server. Hence, the client can specify a "Q- HTTP-Resource-client" header in the BEGIN request of handshake, indicating the relative URI in charge of the server requests when client acts as a server. Example: Q-HTTP-Resource-Client:/example/useragent This URI MUST be relative because user agents may not have an associated domain, in addition to unknown their public IP address. 3.1.1.1. Description of Quality parameters inside SDP The original goal of SDP was designed to announce necessary information for the participants and multicast MBONE (Multicast Backbone) applications. Right now, its use has been extended to the announcement and the negotiation of multimedia sessions. The purpose of SDP in the Q-HTTP context is different because no media parameters are set, therefore the number of media attributes ("m") is always zero. This is because Q-HTTP purpose is not to establish media streams sessions, but monitor a good quality connection, and this quality connection can be used to establish media sessions by other protocols, or for any other purpose. The SDP embedded in the messages is the container of the quality parameters. The included information can comprise all or some of the following parameters, by means of optional session-level attributes: o QoS level for uplink and downlink: specified in the attribute "qos-level". Default values are 0 for both directions. The meaning of each level is out of scope of Q-HTTP, but, in general, a higher level should correspond to a better quality service. o Maximum latency tolerance for uplink and downlink: specified in the attribute "latency", expressed in milliseconds. o Maximum jitter tolerance for uplink and downlink: specified in the attribute "jitter", expressed in milliseconds. Garcia Aranda Expires May 8, 2011 [Page 11] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 o Minimum bandwidth for uplink and downlink: specified in the attribute "bandwidth", expressed in kbps o Maximum packet loss tolerance for uplink and downlink: specified in the attribute "packetloss" expressed in percentage o Flows of data over TCP and UDP ports to be used in uplink and downlink: specified in the attribute "flow" o Measurement procedure and results of quality measurements: specified in the attribute "measurement" This is an example of SDP for Q-HTTP usage. For each attribute two values separated by "/" are involved. These values represent the uplink and downlink values : / . When one or both of these values are empty, it means that there is no constraint on this parameter. v=0 o=q-http-UA 53655765 2353687637 IN IP4 192.0.2.33 s=Q-HTTP i=Q-HTTP parameters t=0 0 a=qos-level:0/0 a=latency:40/35 a=jitter:10/10 a=bandwidth:20/6000 a=packetloss:5/5 a=flow:data downlink TCP/10000-20000 a=flow:control downlink UDP/55000 a=flow:control downlink TCP/55001 a=flow:data uplink TCP/56000 a=flow:control uplink UDP/56000 a=flow:control uplink TCP/56001 a=measurement:procedure default,50/50,75/75,,0 a=measurement:latency 10000/10000 a=measurement:jitter 10000/10000 a=measurement:bandwidth 0/0 a=measurement:packetloss 0/0 Inside the constraints, several "flow" attributes can be defined. The target is to monitor each flow to verify that the quality constraints are met. These flows include the type (uplink or downlink), the protocol (TCP or UDP) (RFC 761 [8] and RFC 768 [9]) and the ports that are going to be used by the application data and, of course, by Garcia Aranda Expires May 8, 2011 [Page 12] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 the control (for quality measurements), because the quality measurements MUST be achieved over the same quality session for each direction. All defined flows will be considered within the same quality profile, which is determined by the qos-level attribute in each direction. During negotiation phase control ports will be used for Q-HTTP messages, and this is the reason to separate application data ports from Q-HTTP control ports, otherwise they could collide. The control should involve two UDP flows (one for uplink and other for downlink) and two TCP flows (one for uplink and other for downlink), but application data could involve many flows, depending on the nature of the application. The initial contact can be achieved at TCP port 80 (for example), but during negotiation phase the control ports (UDP and TCP) will be used instead of the original port used for handshake. The semantics of "downlink port" and "uplink port" is done in reference to destination. Therefore, a downlink port is a port in which client is listening for receiving server messages (and MUST be used as origin port of client responses), and an uplink port is a port in which server is listening incoming messages from client (and MUST be used as origin port of server responses). +------------------------------------------------+ | | | Client Server | | | | downlink port uplink port | | A | | | | | | | +-----------------------------+ | | | | | +------------------------------------------------+ Figure 3 Downlink flow. Garcia Aranda Expires May 8, 2011 [Page 13] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 +------------------------------------------------+ | | | Client Server | | | | downlink port uplink port | | | A | | | | | | +-----------------------------+ | | | | | +------------------------------------------------+ Figure 4 Uplink flow. In addition, measurement parameters are included using the session attribute "measurement". The first measurement parameter is the procedure. By default, Q-HTTP provides a "default" procedure for measurement, but others like RTP/RTCP might be used. In the initial client request a set of measurement procedures can be sent to the server for negotiation (one line MUST be included in SDP for each one). The server will answer with only one line with the chosen procedure. For each procedure, a set of values of parameters can be included in the same attribute line, as in the following example: a=measurement:procedure default,50/50,75/75,5000,0 Where the procedure name is "default" and one parameter is included separated by ",". The meaning of each value depends on the procedure. In the procedure "default", the meaning of these parameters are: o The first parameter is the interval of time (in milliseconds) between PING messages in the negotiation phase. Forward and reverse values are separated by "/". This allows to have two different responsiveness depending on the control resources used in each direction. o The second parameter is the interval of time (in milliseconds) between PING messages in the continuity phase. Forward and reverse values are separated by "/".This allows to have two different responsiveness depending on the control resources used in each direction. Garcia Aranda Expires May 8, 2011 [Page 14] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 o The third parameter is the time used to measure bandwidth during negotiation phase. If not present, a default value of 5000 ms will be assumed. Forward and reverse values are separated by "/". o The fourth parameter indicates the mode for continuity phase (0 means "normal" and 1 means "sliding window"). If not present, normal mode (default value of 0) will be assumed. Quality parameters read by the procedure provide a snapshot of the quality level reached in each stage. Since handshake phase does not make any measurement, this section could be empty or filled with dummy values, except procedure, which is mandatory to start the next protocol phase. 3.1.2. Quality negotiation phase This phase depends on the chosen procedure. The following description corresponds to "default" procedure. The negotiation phase involves iterations of sequences of messages until the quality session is compliant with the minimum quality constraints or until the quality session is closed due to the impossibility to meet the constraints. In order to measure the quality parameters, the client and server can use different mechanisms. This document only describes the "default" mechanism, but others can be used, like RTP/RTCP (RFC 3550 [10]). Measurement of latency and jitter is done calculating the differences in arrival times of packets. This measurement can be achieved with a little bandwidth consumption, whereas bandwidth measurement involves higher bandwidth consumption in both directions (uplink and downlink). Therefore the measurements involve two parts: o Measurement of latencies, jitters and packet loss o Measurement of bandwidths and packet loss Notice that packet loss can be measured in both parts, because the messages used for measure latencies also can be used for packet loss measurement. These two parts are executed sequentially in order to save network resources. If the required latencies and jitters can not be reached, Garcia Aranda Expires May 8, 2011 [Page 15] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 it makes no sense to waste network resources measuring bandwidth. In addition, if the achievement of the required latency and jitter implies upgrading the quality session level, the chance of succeeding in bandwidth measurement without retries is higher, saving network traffic. If the latency and jitter constraints are not empty, the negotiation phase begins with the Measurement of latencies and jitter. Otherwise this stage is skipped. 3.1.2.1. Measurement of latencies and jitters The client starts the negotiation phase sending READY message using the TCP control ports defined in SDP. This READY message includes a specific header "Stage" in which the measurement stage is indicated. In the example, the value 0 means this stage: measurements of latencies, jitters and packet loss. The motivation for this READY message is to synchronize negotiation phases in multiple quality sessions (see 4.2) enabling the possibility to repeat a successful stage. +------------------------------------------------+ | | | Client Server | | | | ------- Q-HTTP READY -----------> | | | | <----- Q-HTTP 200 OK ----------- | | | | | +------------------------------------------------+ Figure 5 Begin of Negotiation phase. Garcia Aranda Expires May 8, 2011 [Page 16] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 Client Request: ========================= READY httpq://www.example.com Q-HTTP/1.0 Stage:0 Session-id: 53655765 User-Agent: qhttp-ua-experimental-1.0 Content-Length: 0 ========================= Server Response: ========================= Q-HTTP/1.0 200 OK Session-id: 53655765 Stage:0 Content-Length: 0 ========================= Just after this, the client MUST send a Q-HTTP message PING using the control flow UDP ports defined in the SDP received at handshake. The downlink port is set as destination and the uplink port is set as origin (according to the example, from client UDP port 56000 to server UDP port 55000). This is an example of the message sent from the client and the server response: Client Request: ========================= PING httpq://www.example.com Q-HTTP/1.0 Session-id: 53655765 Message-id: 0 User-Agent: qhttp-ua-experimental-1.0 Content-Length: 0 ========================= Server Response: ========================= Q-HTTP/1.0 200 OK Session-id: 53655765 Message-id: 0 Content-Length: 0 ========================= The meaning of this method is similar to ICMP PING. Basically the server MUST answer as soon as it receives the message, in order to allow the client to measure the Round trip time (RTT). The RTT is the Garcia Aranda Expires May 8, 2011 [Page 17] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 sum of downlink latency (normally named "reverse latency") and uplink latency (normally named "forward latency"). The client MUST send periodically Q-HTTP PING messages, using always the same UDP control ports and does not need to wait for a response to send the next PING. It simply sends periodically a PING message with a different value of Message-id. Each message is identified by a header "Message-Id". This value is a sequential integer number and MUST start at zero. If this stage is repeated, the initial message-id MUST start again at zero. Optionally the PING request can include a header "Timestamp", with the UTC time in nanoseconds in which the message has been sent. In case the header is present, the server MUST include the header in the response without changing the value. In this phase, the interval between PING messages is defined in the first parameter of the attribute line of SDP where the procedure is specified. In the example, this value is 50 milliseconds (from the client to the server) and 60ms (from the server to the client). a=measurement:procedure default,50/60,50/50,5000,0 A couple of correlated messages (request and response with the same message-id) allow to calculate one sample of RTT. This process could take a few seconds (in the example, five seconds), and after this time, at least 100 samples of RTT MUST be taken by the client. Every time a request message is received by the server, the uplink jitter calculation is updated by the server using the Statistical Jitter value which is calculated on the first 100 packets received using the statistical variance formula: Jitter Statistical = SquareRootOf(SumOf((ElapsedTime[i]- Average)^2)/(ReceivedPacketCount-1)) Hence the client sends a PING periodically with a fixed interval, each value of "elapsed time" (ET) should be very close to this interval. If a PING message is lost, the elapsed time value is doubled, however, this is not an issue because all PING messages are labeled with a Message-Id header. Therefore the receiver can discard this elapsed time value. In order to have the first jitter sample, Garcia Aranda Expires May 8, 2011 [Page 18] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 the server needs to receive 3 PING messages, because each ET is the time between two PINGs and a Jitter needs at least two ET. In order to be able to measure the Jitter in both directions, just after receiving the READY message from the Client, the server MUST begin to do exactly the same, using UDP control ports to send PING messages periodically towards the client. Every time a request PING message is received by the client, the downlink jitter calculation is updated by the client. +------------------------------------------------+ | | | Client Server | | | | --------- Q-HTTP READY -----------> | | <-------- Q-HTTP 200 OK ----------- | | | | --------- Q-HTTP PING ------------> | | <-------- Q-HTTP 200 OK ----------- | | <-------- Q-HTTP PING ------------- | | -------- Q-HTTP 200 OK ----------> | | --------- Q-HTTP PING ------------> | | <-------- Q-HTTP PING ------------- | | --------- Q-HTTP 200 OK ----------> | | <-------- Q-HTTP 200 OK ----------- | | ... | | | +------------------------------------------------+ Figure 6 Latency, jitter and packet loss measurements. After 100 samples the client has the values of RTT and downlink jitter and the server has RTT and uplink jitter. In addition, packet loss is measured in both directions, because Message-id headers are incremented sequentially. Hence, the client knows exactly the number of messages lost from the server to the client, and the server knows the number of packet lost from the client to the server. At this point, the client MUST send a message to the server using TCP control port requesting instructions. This message MUST be sent independently of the used measurement procedure. In the body of the request message the SDP is sent, with updated values of latency, jitter and packet loss. The forward and reverse latencies are unknown Garcia Aranda Expires May 8, 2011 [Page 19] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 and client will assume that the network is symmetric and will assign RTT/2 for uplink and downlink latencies. Client Request: ========================= GET httpq://www.example.com Q-HTTP/1.0 Host: www.example.com User-Agent: qhttp-ua-experimental-1.0 Content-Type: application/sdp Content-Length: 142 v=0 o=q-http-UA 53655765 2353687637 IN IP4 192.0.2.33 s=Q-HTTP i=Q-HTTP parameters t=0 0 a=qos-level:0/0 a=latency:40/35 a=jitter:10/10 a=bandwidth:20/6000 a=packetloss:5/5 a=flow:data downlink TCP/10000-20000 a=flow:control downlink UDP/55000 a=flow:control downlink TCP/55001 a=flow:data uplink TCP/56000 a=flow:control uplink UDP/56000 a=flow:control uplink TCP/56001 a=measurement:procedure default,50/50,75/75,5000,0 a=measurement:latency 40/40 a=measurement:jitter 0/10 a=measurement:bandwidth 0/0 a=measurement:packetloss 0/2 ========================= When the server receives this message,it compares the latency value (RTT/2) with its own measurement, in order to avoid inconsistencies. At this point there are two possibilities o The latency, jitter and packet loss constraints are not reached o The latency, jitter and packet loss constraints are reached Garcia Aranda Expires May 8, 2011 [Page 20] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 3.1.2.1.1. constraints not reached If the measurements do not meet the quality constraints, the server answers with a 412 message (a precondition setting required by the client or server has not been met). Server Answer: ========================= Q-HTTP/1.0 412 latency Date: Mon, 10 Jun 2010 10:00:01 GMT Content-Type: application/sdp Expires: 3000 Cause:downlink_latency Signature: 6ec1ba40e2adf2d783de530ae254acd4f3477ac4 Content-Length: 131 v=0 o=q-http-UA 53655765 2353687637 IN IP4 192.0.2.33 s=Q-HTTP i=Q-HTTP parameters t=0 0 a=qos-level:1/0 a=latency:40/35 a=jitter:10/10 a=bandwidth:20/6000 a=packetloss:5/5 a=flow:data downlink TCP/10000-20000 a=flow:control downlink UDP/55000 a=flow:control downlink TCP/55001 a=flow:data uplink TCP/56000 a=flow:control uplink UDP/56000 a=flow:control uplink TCP/56001 a=measurement:procedure default,50/50,75/75,5000,0 a=measurement:latency 40/40 a=measurement:jitter 20/10 a=measurement:bandwidth 0/0 a=measurement:packetloss 1/2 ========================= In the 412 message, the server may include a different value for "qos-level" SDP session-level attribute, and the measurements done by the client. All these information MUST be protected using the signature header. After a 412 message received by the client, a Q-HTTP message (using TCP control port) with method "QOS-ALERT" is released by the client Garcia Aranda Expires May 8, 2011 [Page 21] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 to acknowledge the SLA violation. Notice that the server signature header is present in the client request, in order to allow an optional integrity validation. If the header "Q-HTTP-policy-server" was included in the server response of the handshake phase, this message MUST be sent to the URI indicated in this header, otherwise the QOS-ALERT message MUST be sent to the server. Client Request: ========================= QOS-ALERT httpq://www.example.com Q-HTTP/1.0 Host: www.example.com User-Agent: qhttp-ua-experimental-1.0 Signature: 6ec1ba40e2adf2d783de530ae254acd4f3477ac4 Content-Type: application/sdp Content-Length: 142 (SDP not shown) ========================= The server answer follows the same syntax as a client request, using a client-server request-response mechanism. Garcia Aranda Expires May 8, 2011 [Page 22] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 Server Answer: ========================= QOS-ALERT httpq://www.example.com Q-HTTP/1.0 Date: Mon, 10 Jun 2010 10:00:01 GMT Content-Type: application/sdp Expires: 3000 Cause: latency Guard-time: 5000 Signature: 6ec1ba40e2adf2d783de530ae254acd4f3477ac4 Content-Length: 131 v=0 o=q-http-UA 53655765 2353687637 IN IP4 192.0.2.33 s=Q-HTTP i=Q-HTTP parameters t=0 0 a=qos-level:1/0 a=latency:40/35 a=jitter:10/10 a=bandwidth:20/6000 a=packetloss:5/5 a=flow:data downlink TCP/10000-20000 a=flow:control downlink UDP/55000 a=flow:control downlink TCP/55001 a=flow:data uplink TCP/56000 a=flow:control uplink UDP/56000 a=flow:control uplink TCP/56001 a=measurement:procedure default,50/50,75/75,5000,0 a=measurement:latency 40/40 a=measurement:jitter 20/10 a=measurement:bandwidth 0/0 a=measurement:packetloss 1/2 ========================= After client receives this answer, client waits for a while indicated in the server message header "Guard-time" (in milliseconds), for example to allow different actions to be carried out by the server. (5 seconds should be enough, but this depends on each case) and begin again the measurement process, starting from the beginning, with the invocation of READY method by the client. The maximum qos-level is 9/9 and if this value is reached without reaching the constraints, the quality session is aborted using the method CANCEL, which is detailed further. Garcia Aranda Expires May 8, 2011 [Page 23] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 If the client does not respect the "Guard-time", and sends the READY message quickly, then the server MUST wait and not answer the READY message until the guard time is elapsed. If during the measurement process some interferences disturb or affect the measurement results, it is better to repeat again the process rather than alerting of an SLA violation. This is always possible by sending current values of parameter "qos-level" without changes, and in this case a header Guard-time can be set to "0". It is a good practice to repeat the measurements before reporting a violation. 3.1.2.1.2. Constraints not reached with Policy server involved If during handshake phase the optional header Q-HTTP-policy-server is included in the server response, the QOS-ALERT message MUST be sent to the policy server, which should implement all or some of these features (but not exclusive to): o Client and server validation in terms of SLA. o Authentication (Signature validation) and security (block malicious clients) o Policy rules ( following rules are only examples): - Maximum quality level allowed for the ACP - Time bands allowed for provide quality sessions for the ACP - Number of simultaneous quality sessions allowed - Maximum time used by quality sessions allowed - Etc. With policy server, the QOS-ALERT message sent by the client MUST contain the URIs of the server and the client to be contacted later by the policy server. Therefore the following headers MUST be included in the client request: "Q-HTTP-Resource-server" and "Q-HTTP- Resource-client" Depending on the results of the operations achieved by polity server, the client could receive different types of errors or CANCEL messages. Garcia Aranda Expires May 8, 2011 [Page 24] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 The flows of messages in this case are in the following figure: +------------------------------------------------+ | | | Client policy Server | | | | --- QOS-ALERT -----> | | <-- 100 trying ----- | | | | ---- QOS-ALERT ----> | | <--- QOS-ALERT ----- | | <--- QOS-ALERT ----- | | | +------------------------------------------------+ Figure 7 Policy server. If the validation or authentication of the QOS-ALERT operation fails, the policy will send a CANCEL operation to the client without contacting the server. If any of the policy rules fail, the server will send a 6XX error to the client, indicating the rule which is not satisfied. Only if the validation, authentication and policy checking are successful, the server is contacted by the policy server and the QOS- ALERT message is forwarded to it. 3.1.2.1.3. Constraints reached When latency and jitter measurements match the constraints, the server answer should be 200 OK: Server Answer: ========================= Q-HTTP/1.0 200 OK Date: Mon, 10 Jun 2010 10:00:01 GMT Content-Type: application/sdp Expires: 3000 Signature: 6ec1ba40e2adf2d783de530ae254acd4f3477ac4 Content-Length: 131 (SDP not shown) ========================= Garcia Aranda Expires May 8, 2011 [Page 25] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 It means that the client and the server are ready for bandwidth and packet loss measurement. If the bandwidth constraints are not empty, the negotiation phase continues with the Measurement of bandwidth and packet loss. Otherwise this stage is skipped. Garcia Aranda Expires May 8, 2011 [Page 26] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 +------------------------------------------------+ | | | Client Server | | | | --------- Q-HTTP READY -----------> | | <-------- Q-HTTP 200 OK ----------- | | | | --------- Q-HTTP PING ------------> | | <-------- Q-HTTP 200 OK ----------- | | <-------- Q-HTTP PING ------------- | | --------- Q-HTTP 200 OK ----------> | | --------- Q-HTTP PING ------------> | | <-------- Q-HTTP PING ------------- | | <-------- Q-HTTP 200 OK ----------- | | --------- Q-HTTP 200 OK ----------> | | ... | | --------- Q-HTTP GET -------------> | | <-------- Q-HTTP 412 -------------- | | --------- Q-HTTP QOS-ALERT -------> | | <-------- Q-HTTP QOS-ALERT -------- | | (delay) | | --------- Q-HTTP PING ------------> | | <-------- Q-HTTP PING ------------- | | <-------- Q-HTTP 200 OK ----------- | | --------- Q-HTTP 200 OK ----------> | | ... | | --------- Q-HTTP GET -------------> | | <-------- Q-HTTP 412 -------------- | | --------- Q-HTTP QOS-ALERT -------> | | <-------- Q-HTTP QOS-ALERT -------- | | (delay) | | --------- Q-HTTP PING ------------> | | <-------- Q-HTTP PING ------------- | | <-------- Q-HTTP 200 OK ----------- | | --------- Q-HTTP 200 OK ----------> | | ... | | --------- Q-HTTP GET -------------> | | <-------- Q-HTTP 200 OK ----------- | | | +------------------------------------------------+ Figure 8 Latency and jitter measurements with final success Garcia Aranda Expires May 8, 2011 [Page 27] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 3.1.2.2. Measurement of bandwidth and packet loss This stage begins in the same way as the previous one, sending a READY message over TCP control ports. This READY message includes a specific header "Stage" in which the measurement stage is indicated. In the example, the value 1 means this stage: measurements of bandwidth and packet loss. +------------------------------------------------+ | | | Client Server | | | | --------- Q-HTTP READY -----------> | | <-------- Q-HTTP 200 OK ----------- | | | +------------------------------------------------+ Figure 9 Starting bandwidth and packet loss measurement Client Request: ========================= READY httpq://www.example.com Q-HTTP/1.0 User-Agent: qhttp-ua-experimental-1.0 Stage:1 Session-id: 53655765 Content-Length: 0 ========================= Server Response: ========================= Q-HTTP/1.0 200 OK Session-id: 53655765 Stage:1 Content-Length: 0 ========================= Just after receiving the 200 OK, both client and the server MUST start sending messages simultaneously using the UDP control ports, at the needed rate to reach the bandwidth constraint in each direction using messages of 1 Kbyte length. The messages are sent during a period of time defined in the SDP. This time is the third parameter of procedure "default", in milliseconds. If this parameter is not present, a value of 5 seconds will be used by default. a=measurement:procedure default,50/50,75/75,5000,0 Garcia Aranda Expires May 8, 2011 [Page 28] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 +------------------------------------------------+ | Rate | | A | | | | |rate downlink-|-------------------+ <-- traffic | | | | sent by | | | | server | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | rate uplink-|-------------------+ <-- traffic | | | | sent by | | | | client | | | | | | | | | | |---|---|---|---|---|----> time | | 0 1 2 3 4 5 (sec.) | | | +------------------------------------------------+ Figure 10 Bandwidth and packet loss measurements. The goal of this phase is not to measure the internet connection bandwidth connection but to measure if the quality constraints can be reached or not. This is the reason for not sending more bit rate than needed. All messages to be sent MUST be 1 kilobyte length (hence the size of the body depends on the size of included headers), and include a Message-id header with a sequential number which starts at 0. If the stage is repeated, the values MUST start again at zero. Examples: Garcia Aranda Expires May 8, 2011 [Page 29] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 Client message: ========================= DATA httpq://www.example.com Q-HTTP/1.0 User-Agent: qhttp-ua-experimental-1.0 Session-id: 53655765 Message-id: 0 Content-Type: text Content-Length: XXXX aaaaaaaaaaaaa ( to complete 1024 bytes packet length) ========================= The messages MUST NOT be answered, but only sent. The client will send packets to the server in order to allow server measure client bandwidth, and the server will do the same towards the client. The packets have a message-Id to be aware of the packet loss at reception. The value of message-Id will start at cero and will be incremented by 1 for each message. server message: ========================= DATA httpq://www.example.com Q-HTTP/1.0 Session-id: 53655765 Message-id: 0 Content-Type: text Content-Length: 1024 aaaaaaaaaaaaa ( to complete 1024 bytes packet length) ========================= After a 5 seconds measurements the client has a collection of server messages and may calculate the packet loss and downlink bandwidth received. At the other side, the server has the uplink bandwidth and packet loss. Client MUST send a GET message to the server using the TCP control port including the SDP data filled up with the measured downlink bandwidth and packet loss. Garcia Aranda Expires May 8, 2011 [Page 30] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 Client Request: ========================= GET httpq://www.example.com Q-HTTP/1.0 Host: www.example.com User-Agent: qhttp-ua-experimental-1.0 Session-id: 53655765 Content-Type: application/sdp Content-Length: 142 v=0 o=q-http-UA 53655765 2353687637 IN IP4 192.0.2.33 s=Q-HTTP i=Q-HTTP parameters t=0 0 a=qos-level:1/1 a=latency:40/35 a=jitter:10/10 a=bandwidth:20/6000 a=packetloss:5/5 a=flow:data downlink TCP/10000-20000 a=flow:control downlink UDP/55000 a=flow:control downlink TCP/55001 a=flow:data uplink TCP/56000 a=flow:control uplink UDP/56000 a=flow:control uplink TCP/56001 a=measurement:procedure default,50/50,50/50,5000,0 a=measurement:latency 30/30 a=measurement:jitter 6/4 a=measurement:bandwidth 0/4000 a=measurement:packetloss 0/3 ============================== At this point there are two possibilities: o The bandwidth and packet loss constraints are not reached in any or both directions. o The bandwidth and packet loss constraints are reached in both directions. 3.1.2.2.1. constraints not reached If the measurements does not reach the quality constraints, the server answers with a 412 message (a precondition setting required by the client or server has not been met). Otherwise returns 200 OK. Garcia Aranda Expires May 8, 2011 [Page 31] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 In the 412 message, the server may include a different value for "qos-level" SDP session-level attribute, and the measurements of bandwidth and packet loss in both directions. All these information MUST be protected using the signature header. Server Answer: ========================= Q-HTTP/1.0 412 downlink_bandwidth Date: Mon, 10 Jun 2010 10:00:01 GMT Content-Type: application/sdp Expires: 3000 Cause:downlink_bandwidth Signature: 6ec1ba40e2adf2d783de530ae254acd4f3477ac4 Content-Length: 131 v=0 o=q-http-UA 53655765 2353687637 IN IP4 192.0.2.33 s=Q-HTTP i=Q-HTTP parameters t=0 0 a=qos-level:1/2 a=latency:40/35 a=jitter:10/10 a=bandwidth:20/6000 a=packetloss:5/5 a=flow:data downlink TCP/10000-20000 a=flow:control downlink UDP/55000 a=flow:control downlink TCP/55001 a=flow:data uplink TCP/56000 a=flow:control uplink UDP/56000 a=flow:control uplink TCP/56001 a=measurement:procedure default,50/50,50/50,5000,0 a=measurement:latency 30/30 a=measurement:jitter 6/4 a=measurement:bandwidth 200/4000 a=measurement:packetloss 2/3 ========================= After a 412 message client MUST send a Q-HTTP message (using TCP control port) with method "QOS-ALERT" to acknowledge the SLA violation. Notice that the server signature header is present in the client request, in order to allow integrity validation. Garcia Aranda Expires May 8, 2011 [Page 32] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 Client Request: ========================= QOS-ALERT httpq://www.example.com Q-HTTP/1.0 Host: www.example.com User-Agent: qhttp-ua-experimental-1.0 Signature: 6ec1ba40e2adf2d783de530ae254acd4f3477ac4 Content-Type: application/sdp Content-Length: 142 v=0 o=q-http-UA 53655765 2353687637 IN IP4 192.0.2.33 s=Q-HTTP i=Q-HTTP parameters t=0 0 a=qos-level:1/2 a=latency:40/35 a=jitter:10/10 a=bandwidth:20/6000 a=packetloss:5/5 a=flow:data downlink TCP/10000-20000 a=flow:control downlink UDP/55000 a=flow:control downlink TCP/55001 a=flow:data uplink TCP/56000 a=flow:control uplink UDP/56000 a=flow:control uplink TCP/56001 a=measurement:procedure default,50/50,50/50,5000,0 a=measurement:latency 30/30 a=measurement:jitter 6/4 a=measurement:bandwidth 200/4000 a=measurement:packetloss 2/3 ========================= The server answer follows the same syntax as a client request. Garcia Aranda Expires May 8, 2011 [Page 33] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 Server Answer: ========================= QOS-ALERT httpq://www.example.com Q-HTTP/1.0 Date: Mon, 10 Jun 2010 10:00:01 GMT Content-Type: application/sdp Expires: 3000 Cause: latency Signature: 6ec1ba40e2adf2d783de530ae254acd4f3477ac4 Content-Length: 131 v=0 o=q-http-UA 53655765 2353687637 IN IP4 192.0.2.33 s=Q-HTTP i=Q-HTTP parameters t=0 0 a=qos-level:1/2 a=latency:40/35 a=jitter:10/10 a=bandwidth:20/6000 a=packetloss:5/5 a=flow:data downlink TCP/10000-20000 a=flow:control downlink UDP/55000 a=flow:control downlink TCP/55001 a=flow:data uplink TCP/56000 a=flow:control uplink UDP/56000 a=flow:control uplink TCP/56001 a=measurement:procedure default,50/50,50/50,5000,0 a=measurement:latency 30/30 a=measurement:jitter 6/4 a=measurement:bandwidth 200/4000 a=measurement:packetloss 2/3 ========================= Once client receives this answer, client and server wait for a while indicated in the server message header "Guard-time" (in milliseconds), for example to allow different actions to be carried out by the server (5 seconds should be enough, but this depends on each case) and begin again the measurement process (bandwidth and packet loss), starting with a READY message indicating the current stage (1). The maximum qos-level is 9/9 and if this value is reached without matching the constraints, the quality session is aborted using the method CANCEL, which is detailed further in this document. Garcia Aranda Expires May 8, 2011 [Page 34] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 3.1.2.2.2. Constraints not reached with Policy server involved If during handshake phase the optional header Q-HTTP-policy-server is included in the server response, the QOS-ALERT message MUST be sent to the policy server. The involved messages and operations are described in 2.1.2.1.2 3.1.2.2.3. Constraints reached When measurements match the constraints, the server's answer should be 200 OK, and MUST include the URI for trigger the application using an optional header "Trigger-URI" Server Answer: ========================= Q-HTTP/1.0 200 OK Date: Mon, 10 Jun 2010 10:00:01 GMT Trigger-URI:http://www.example.com/app_start Expires: 3000 Content-Type: application/sdp Signature: 6ec1ba40e2adf2d783de530ae254acd4f3477ac4 Content-Length: 131 (SDP not shown) ========================= It means that client and server are ready to start the application. Garcia Aranda Expires May 8, 2011 [Page 35] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 +------------------------------------------------+ | | | Client Server | | | | <-------- (DATA packets) ------------> | | ... | | --------- Q-HTTP GET ----------------> | | <-------- Q-HTTP 412 ----------------- | | ---- Q-HTTP QOS-ALERT ---------------> | | <--- Q-HTTP QOS-ALERT ---------------- | | (delay) | | --------- Q-HTTP READY --------------> | | <-------- Q-HTTP 200 OK -------------- | | <-------- (DATA packets) ------------> | | ... | | --------- Q-HTTP GET ----------------> | | <-------- Q-HTTP 412 ----------------- | | ---- Q-HTTP QOS-ALERT ---------------> | | <--- Q-HTTP QOS-ALERT ---------------- | | (delay) | | --------- Q-HTTP READY---------------> | | <-------- Q-HTTP 200 OK -------------- | | <-------- (DATA packets) ------------> | | ... | | --------- Q-HTTP GET ----------------> | | <-------- Q-HTTP 200 OK--------------- | | | | | | | +------------------------------------------------+ Figure 11 Bandwidth & packet loss measurement with success. 3.1.2.3. Qos Level out of range If the qos-level has reached the maximum value for downlink or uplink without matching the constraints, then a CANCEL Q-HTTP message MUST be sent in order to release the session. This message MUST be sent using the control TCP port by client and the server MUST answer CANCEL too. Otherwise, the expiration time cancels the session at server side. Garcia Aranda Expires May 8, 2011 [Page 36] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 Client Request: ========================= CANCEL httpq://www.example.com Q-HTTP/1.0 Host: www.example.com User-Agent: qhttp-ua-experimental-1.0 Signature: 6ec1ba40e2adf2d783de530ae254acd4f3477ac4 Content-Type: application/sdp Content-Length: 142 (SDP not shown) ========================= Server Answer: ========================= CANCEL httpq://www.example.com Q-HTTP/1.0 Date: Mon, 10 Jun 2010 10:00:01 GMT Expires: 0 Content-Type: application/sdp Signature: 6ec1ba40e2adf2d783de530ae254acd4f3477ac4 Content-Length: 131 (SDP not shown) ========================= The server answer follows the same syntax as a client request. Garcia Aranda Expires May 8, 2011 [Page 37] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 +------------------------------------------------+ | | | Client Server | | | | <-------- (measurements) ------------> | | | | --------- Q-HTTP GET ----------------> | | <-------- Q-HTTP 412 ----------------- | | ---- Q-HTTP QOS-ALERT ---------------> | | <--- Q-HTTP QOS-ALERT --------------- | | --------- Q-HTTP READY --------------> | | <-------- Q-HTTP 200 OK -------------- | | | | <-------- (measurements) ------------> | | | | --------- Q-HTTP GET ----------------> | | <-------- Q-HTTP 412 ----------------- | | --------- Q-HTTP CANCEL -------------> | | <-------- Q-HTTP CANCEL -------------- | | | | | +------------------------------------------------+ Figure 12 Negotiation phase failed. 3.1.2.4. Qos Level increments without changes in network behaviour If the qos-level has not reached the maximum value (9) but after 3 QOS-ALERT messages (with increments in qos-level) the network remains with the same quality values, the client and the server MUST understand that the network can not reach the desired quality and will abort the session in order to save resources (time and traffic). To do that, client MUST sent a CANCEL message and the server MUST answer with a CANCEL message too. If the client does not send a CANCEL message but any other, the server MUST answer with a CANCEL message. 3.1.2.5. Trigger an application in combination with HTTP When the negotiation phase is successful, an optional simple mechanism, based on http, is defined to trigger the application. The application may be triggered using an URI, by means of an http request, just after negotiation success. The URI MUST be specified in the Q-http header "Trigger-URI". Other mechanisms, such as including a "Location" header in the Q-http message, to force redirection is Garcia Aranda Expires May 8, 2011 [Page 38] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 not recommended because these mechanisms are achieved without parsing the body of the message. Example of use +------------------------------------------------+ | | | Client Server | | | | --------- HTTP GET ----------------> | | <-------- redirect to httpq ---------- | | | | ------- Q-HTTP BEGIN ----------------> | | | | (Handshake Phase) | | (Negotiation Phase) | | | | <---- Q-HTTP 200 OK with trigger URI-- | | | | --------- HTTP GET ----------------> | | | | (Application starts) | | | +------------------------------------------------+ Figure 13 Trigger the application using HTTP URI In the example, an integration of http and Q-http is shown. First, the client contacts the server using http, a redirection to a Q-http URI is achieved and the User Agent starts the Q-http handshake phase. After negotiation phase succeeds, the client trigger the application using the URI indicated in the Q-http 200 OK message. 3.1.3. Continuity phase During negotiation phase the latency, jitter, bandwidth and packet loss can be measured, but during continuity phase bandwidth will not be measured because bandwidth measurements may disturb application performance. This phase is supposed to be executed at the same time as the real time application is being used. In the default measurement procedure, two working modes are defined for this phase (normal and sliding window). The details of working Garcia Aranda Expires May 8, 2011 [Page 39] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 modes are procedure dependant, and this draft only covers the default procedure. 3.1.3.1. Normal mode The server can force the use of normal mode by setting the fourth parameter of "procedure" SDP attribute to 0. If this parameter is not set, the default value is assumed (cero), and normal mode will be used. Example: a=measurement:procedure default,50/50,50/50,5000,0 Considering that network conditions can change, periodically the client may re-executes the negotiation phase. The maximum interval expected to restart the negotiation phase is indicated in the Q-HTTP Expires header. However, the measurements can be achieved periodically with a smaller period of time than "Expires" header value, in order to make sure that the communication matches the constraints. In intense interactive applications, like arcade videogames, the period to repeat the measurements may be very small (even cero), in order to measure continuously the quality and assure the best reaction time. To reach the best reaction time, the use of sliding window mode is recommended. To start the continuity phase, the client sends a Q-HTTP READY method, using the TCP control port, exactly the same as Negotiation, indicating the new Stage header value for continuity phase (value 2). Garcia Aranda Expires May 8, 2011 [Page 40] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 Client Request: ========================= READY httpq://www.example.com Q-HTTP/1.0 User-Agent: qhttp-ua-experimental-1.0 Stage:2 Session-id: 53655765 Content-Length: 0 ========================= Server Response: ========================= Q-HTTP/1.0 200 OK Session-id: 53655765 Stage:2 Content-Length: 0 ========================= After these messages starts latency, jitter and packet loss measurement, taking care of bandwidth usage. If the default measurement method is being used, it is recommended to use a larger interval for PING messages, but the same number of samples will be taken to check quality. The goal of increment the interval of PING messages is to minimize the load of the server which would be running lots of connections in parallel. The process is the same as described in the negotiation phase. The difference is the time between samples, because the bandwidth usage MUST be protected. The interval used for this phase is indicated in the second parameter of the attribute line for the procedure. In this example, the interval is 75 milliseconds. a=measurement:procedure default,50/50,75/75,5000,0 A value larger than used in negotiation phase is recommended, but not mandatory. Garcia Aranda Expires May 8, 2011 [Page 41] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 +------------------------------------------------+ | | | Client Server | | | | | | --------- Q-HTTP READY -----------> | | <-------- Q-HTTP 200 OK ----------- | | --------- Q-HTTP PING ------------> | | <-------- Q-HTTP 200 OK ----------- | | <-------- Q-HTTP PING ------------- | | -------- Q-HTTP 200 OK ----------> | | --------- Q-HTTP PING ------------> | | <-------- Q-HTTP PING ------------- | | --------- Q-HTTP 200 OK ----------> | | <-------- Q-HTTP 200 OK ----------- | | ... | | --------- Q-HTTP GET -------------> | | <-------- Q-HTTP 412 -------------- | | --------- Q-HTTP QOS-ALERT -------> | | <-------- Q-HTTP QOS-ALERT -------- | | (delay) | | --------- Q-HTTP READY -----------> | | <-------- Q-HTTP 200 OK ----------- | | --------- Q-HTTP PING ------------> | | <-------- Q-HTTP 200 OK ----------- | | <-------- Q-HTTP PING ------------- | | -------- Q-HTTP 200 OK ----------> | | --------- Q-HTTP PING ------------> | | <-------- Q-HTTP PING ------------- | | --------- Q-HTTP 200 OK ----------> | | <-------- Q-HTTP 200 OK ----------- | | ... | | --------- Q-HTTP GET -------------> | | <-------- Q-HTTP 200 OK ----------- | | | +------------------------------------------------+ Figure 14 Continuity. 3.1.3.2. Sliding window mode In order to improve the reaction time when network conditions degrade quickly, the server can force the use of sliding window mode by setting the fourth parameter of "procedure" SDP attribute to 1. Example: Garcia Aranda Expires May 8, 2011 [Page 42] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 a=measurement:procedure default,50/50,50/50,5000,1 The sliding window mode applies a sliding window of 100 samples instead cycles of 100 samples. In the sliding window mode, PING messages are sent continuously (by client and server) and when Message-id header reach the value of 100, client MUST NOT send a GET message for instructions, but continues sending PING messages with Message-id header starting again at zero. When the server PING Message-id header reaches 100, do the same, starting again at zero. On the client side, the measured values of downlink jitter, downlink packet loss and latency are calculated using the last samples, discarding older ones, in a sliding window schema. +------------------------------------------------+ | | | 55 56 57 . . . 98 99 100 0 1 2 . . . 55 56 | | A A | | | | | | +---------------------------------+ | | | +------------------------------------------------+ Figure 15 Sliding samples window Only when the client detects that the measured values (downlink jitter, downlink packet loss and latency) are not reaching the constraints, send a GET message to the server. When the server receives the Q-HTTP GET message, it stops sending PING packets and answer the GET request. If a 412 message is answered, then a QOS-ALERT will be requested by client, exactly in the same way as described in normal mode. On the other hand, if the server detects that the measured values (uplink jitter, uplink packet loss and latency) are not reaching the constraints, it MUST choose between the following alternatives: o The server stops sending PING messages to the client. In this case the client MUST notice this lack of PING messages by using a timeout at reception, and it reacts stopping the sending of PING messages and sends a GET message for instructions, exactly in the same way as described in normal mode. Garcia Aranda Expires May 8, 2011 [Page 43] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 o It continues sending PING messages but all of them with Message-id set to -1 till a client GET message is received. Then the server stops sending PING messages and answers the GET request with the corresponding 412 error, exactly in the same way as described in normal mode. Client reacts when it receives a PING with Message-id header set to -1, sending this GET request. This behaviour allows the shortest reaction time under degradation of network conditions. Both alternatives MUST be implemented by the Q-HTTP client. 3.2. Dynamic constraints and flows Depending on the nature of the application, the constraints to be reached can evolve, changing some of all constraint values in both directions. This possibility MUST be supported by the client. When the server sends a SDP embedded into a error message (200 OK, or 412, etc), the client MUST assume all the new values of the received SDP. The dynamic changes on the constraints can be the result of these two possibilities: o If the application communicates with the Q-HTTP server to change constraints. In this case the application requirements can evolve and Q-HTTP server will be aware of them. o If the application uses TCP flows. In that case, in order to guarantee a constant throughput, the nature of TCP behavior forces the use of a composite constraint function which depends on RTT, packet loss and window control mechanism implemented in each TCP stack. TCP throughput can be less than actual bandwidth in particular if the Bandwidth-Delay Product (BDP) is large or if a network suffers from a high packet loss rate. In both cases, TCP's congestion control algorithms may result in a suboptimal performance. Different TCP congestion control implementations like Reno [14], High Speed TCP (RFC 3649 [15]), CUBIC [16], Compound TCP (CTCP [17]), etc. reach different throughputs under the same network conditions of RTT and packetloss. In all cases, depending on RTT measured value, Q-HTTP server could change dynamically the packetloss constraints (defined in SDP) in order to make possible to reach a required throughput or viceversa (use packetloss measurement to change dynamically latency constraints). Garcia Aranda Expires May 8, 2011 [Page 44] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 A general guideline to calculate the packetloss constraint and RTT constraint consists in approximating the throughput using a simplified formula which should take into account the TCP stack implementation of the receiver, in addition to RTT and packet loss: Th= Function( RTT, packet loss, ...) Then, depending on RTT measured values, set dynamically the packetloss constraint. It is possible to calculate easily a worst-case boundary for the Reno algorithm which should ensure for all algorithms that the target throughput is actually achieved. Except that, high-speed algorithms will then have even a larger throughput, if more bandwidth is available. For Reno algorithm, it may be used the Mathis' formula [15] for the upper bound on the throughput : Th <= (MSS/RTT)*(1 / sqrt{p}) In absence of packet loss, a practical limit for TCP throughput is the receiver_window_size divided by the round-trip time. However, if the TCP implementation uses window scale option, this limit can reach the available bandwidth value. 3.3. QoS-level downgrade operation During the continuity phase might be desirable to downgrade the current QoS-level SDP parameter. The strategy to carry out downgrades must include the possibility to exclude specific data flows from SDP dynamically. A Q-HTTP client MUST allow this kind of SDP modifications by server. Periodically (each several minutes, depending on the implementation) server could force a QOS-ALERT, in which the level is downgraded for control flows, excluding application data flows from the embedded SDP of that request. To set the new SDP, the server MUST include the modified SDP in the 412 error message. This mechanism allows to measure at lower levels of quality while application flows continue using a higher qos level value Garcia Aranda Expires May 8, 2011 [Page 45] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 o If the measurements in the lower level meet the constraints, then a new QOS-ALERT to this lower qos-level can be forced by server, in which the SDP includes the application data flows in addition to control flows. o If the measurements in the lower level do not meet the constraints, then a new QOS-ALERT to the previous qos-level MUST be forced by the server, in which the SDP includes only the control flows. +------------------------------------------------+ | | | qos-level | | A | | | | | 4| | | | | | 3| +------+ | | | | | | | 2| +----+ +----+ +--- | | | | | | | | 1| +----+ +-----+ | | | | | | 0+---+---------------------------------> time | | | +------------------------------------------------+ Figure 16 Possible evolution of qos-level This mechanism avoids the risk of disturbing the application, while the measurements are being run in lower levels. However, this optimization of resources is optional, and MUST be used carefully. The chosen period to measure a lower qos level is implementation dependant. Therefore it is not included as a measurement procedure parameter. It is recommended to use a large value, such as 20 minutes. 3.4. Sanity check of Quality sessions A session may finish by several reasons (client shutdown, client CANCEL request, constraints not reached, etc), and any session finished MUST release the assigned resources. In order to release the assigned server resources for the session, the header "Expires" indicate the maximum interval of time that a client can wait to repeat the continuity phase (in normal mode). Garcia Aranda Expires May 8, 2011 [Page 46] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 4. Q-HTTP messages Q-HTTP is a text-based protocol and uses the UTF-8 charset (RFC 3629 [11]). A Q-HTTP message is either a request or a response. Both Request and Response messages use the basic format of Internet Message Format (RFC 5322 [12]). Both types of messages consist of a start-line, one or more header fields, an empty line indicating the end of the header fields, and an optional message-body. generic-message = start-line *message-header CRLF [ message-body ] start-line = Request-Line / Status-Line The start-line, each message-header line, and the empty line MUST be terminated by a carriage-return line-feed sequence (CRLF). Note that the empty line MUST be present even if the message-body is not. Much of Q-HTTP's messages and header field syntax are identical to HTTP/1.1. However, Q-HTTP is not an extension of HTTP. 4.1. Requests Q-HTTP requests are distinguished by having a Request-Line for a start-line. A Request-Line contains a method name, a Request-URI , and the protocol version separated by a single space (SP) character. The Request-Line ends with CRLF. No CR or LF are allowed except in the end-of-line CRLF sequence. No linear whitespace (LWS) is allowed in any of the elements. Request-Line = Method SP Request-URI SP Q-HTTP-Version CRLF Method: This specification defines five methods: GET for get information and send quality reports, PING and DATA for quality measurements purpose, CANCEL for terminating sessions, and QOS-ALERT for querying ISPs for quality upgrades. Garcia Aranda Expires May 8, 2011 [Page 47] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 Request-URI: The Request-URI is a Q-http URI (RFC 2396 )as described in 2.2.1 It Normally indicates the user or service to which this request is being addressed, but in the Q-http context, there are some methods which URI only reflects the service on the server side, but nor more. This is the case of method QOS- ALERT, because the real address of a QoS upgrade request is the network, and therefore in this case the URI only reflects the server address. In addition the method CANCEL has the same treatment, and in the methods ECHO and DATA invoked by server to the client the meaning of the URI is only the URI of the service, but not the destination of the request. The Request- URI MUST NOT contain unescaped spaces or control characters and MUST NOT be enclosed in "<>". Q-HTTP-Version: Both request and response messages include the version of Q-HTTP in use. To be compliant with this specification, applications sending Q-HTTP messages MUST include a Q-HTTP-Version of "Q-HTTP/1.0". The Q-HTTP-Version string is case-insensitive, but implementations MUST send upper-case. Unlike HTTP/1.1, Q-HTTP treats the version number as a literal string. In practice, this should make no difference. 4.2. Responses In Q-HTTP there are 2 types of responses: o Responses that follow the same syntax rules of a request: these cases are for methods which suggest actions, as QOS-ALERT and CANCEL. After a successful negotiation phase, instead of using a 2xx response code, a request is generated as response message. However, these methods can have a conventional answer if an error is detected. o Conventional responses, where the server is answering to a previous client request. Q-HTTP conventional responses are distinguished from requests by having a Status-Line as their start-line. A Status-Line consists of the protocol version followed by a numeric Status-Code and its associated textual phrase, with each element separated by a single SP character. No CR or LF is allowed except in the final CRLF sequence. Garcia Aranda Expires May 8, 2011 [Page 48] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 Status-Line = Q-HTTP-Version SP Status-Code SP Reason-Phrase CRLF The Status-Code is a 3-digit integer result code that indicates the outcome of an attempt to understand and satisfy a request. The Reason-Phrase is intended to give a short textual description of the Status-Code. The Status-Code is intended for use by automata, whereas the Reason-Phrase is intended for the human user. A client is not required to examine or display the Reason-Phrase. While this specification suggests specific wording for the reason phrase, implementations MAY choose other text, for example, in the language indicated in the Accept-Language header field of the request. The first digit of the Status-Code defines the class of response. The last two digits do not have any categorization role. For this reason, any response with a status code between 100 and 199 is referred to as a "1xx response", any response with a status code between 200 and 299 as a "2xx response", and so on. Q-HTTP/1.0 allows following values for the first digit: 1xx: Provisional -- request received, continuing to process the request; 2xx: Success -- the action was successfully received, understood, and accepted; 3xx: Redirection -- further action needs to be taken in order to complete the request; 4xx: Client Error -- the request contains bad syntax or cannot be fulfilled at this server; 5xx: Server Error -- the server failed to fulfill an apparently valid request; 6xx: Global Failure -- the request cannot be fulfilled at any server. The status codes are the same described in HTTP (RFC 2616 [1]). In the same way as HTTP, Q-HTTP applications are not required to understand the meaning of all registered status codes, though such understanding is obviously desirable. However, applications MUST understand the class of any status code, as indicated by the first Garcia Aranda Expires May 8, 2011 [Page 49] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 digit, and treat any unrecognized response as being equivalent to the x00 status code of that class 4.3. Header Fields Q-HTTP header fields are identical to HTTP header fields in both syntax and semantics. Some header fields only make sense in requests or responses. These are called request header fields and response header fields, respectively. If a header field appears in a message not matching its category (such as a request header field in a response), it MUST be ignored. 4.3.1. Specific Q-HTTP Request Header Fields In addition to HTTP header fields, these are the specific Q-HTTP request header fields o Session-id: the value for this header is the same session id used in SDP and is assigned by server. The messages without SDP MUST include this header. If a message has SDP, the header is optional. The method of allocation is up to the creating tool, but it has been suggested that a Network Time Protocol (NTP) timestamp be used to ensure uniqueness. o Message-id: sequential integer number assigned to PING and DATA messages. o Timestamp: UTC time in nanoseconds. Indicates the time in which the request was sent. o Signature: this header contains a digital signature that can be used by the network to validate the SDP. The signature is always generated by the server. It is optional. o Q-HTTP-Resource-client: this optional header contains the relative URI in charge of this session. In The case of being included, it MUST appear in the GET request of handshake phase. This URI MUST be invoked by the server in all later requests. It is optional, but it should be present, it becomes mandatory for the counterpart. This URI MUST be relative because user agents can not have associated domain, in addition to ignore their public IP address. Garcia Aranda Expires May 8, 2011 [Page 50] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 4.3.2. Specific Q-HTTP Response Header Fields o Expires: the purpose is to provide a sanity check and allow the server to close inactive sessions. If the client does not send a new request before the expiration time, the server can close the session. The value MUST be an integer and the measurement unit are milliseconds. o Guard-time: A time interval left vacant (i.e., during which no data is sent) during the quality session. The guard time provides a safety margin before re-starting each measurement process when a QOS-ALERT has been raised. This header is optional in all messages but mandatory in the QOS-ALERT sent by the server. o Message-id: same meaning as Request Header Fields o Timestamp: UTC time in nanoseconds. Indicates the time in which the request was sent. If the server (or a client) receives a Timestamp header in a request, MUST include the same header with the same value in the response. The purpose of this header is simplify the RTT calculation. o Signature: same meaning as Request Header Fields o Q-HTTP-Resource-server: this optional header contains the URI in charge of this session. In case of being included, it MUST appear in the response of handshake phase. This URI MUST be invoked by the client in all later requests. It is optional, but it should be present, it becomes mandatory for the counterpart. o Q-HTTP-policy-server: this optional header contains the URI towards the client and MUST send the QOS-ALERT messages. In case this header is present, the header Q-HTTP-Resource-server is mandatory, and MUST be included in the QOS-ALERT messages sent by the client to the policy server. In addition, the QOS- ALERT sent to the policy server MUST contain the header Q-HTTP- Resource-client 4.4. Bodies Requests, including new requests defined in extensions to this specification, MAY contain message bodies unless otherwise noted. The interpretation of the body depends on the request method. Garcia Aranda Expires May 8, 2011 [Page 51] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 For response messages, the request method and the response status code determine the type and interpretation of any message body. All responses MAY include a body. The Internet media type of the message body MUST be given by the Content-Type header field. 4.4.1. Encoding The body MUST NOT be either encoded or compressed. This mechanism is valid for other protocols such as HTTP and SIP (RFC 3261 [13]), but a compression/coding scheme will limit certain logical implementations of the way the request is parsed, thus, making the protocol concept more implementation dependant. In addition, bandwidth calculation may not be valid if compression is used. Therefore, the HTTP request header "Accept-Encoding" can not be used in Q-HTTP with different values than "identity" and if it is present in a request, the server MUST ignore it. In addition, the response header "Content-Encoding" is optional, but if present, the unique permitted value is "identity". The body length in bytes is provided by the Content-Length header field. The "chunked" transfer encoding of HTTP/1.1 MUST NOT be used for Q-HTTP (Note: The chunked encoding modifies the body of a message in order to transfer it as a series of chunks, each one with its own size indicator.) 5. General User Agent behavior. 5.1. Roles In order to allow peer to peer applications, a Q-HTTP User Agent (UA) MUST be able to assume both client and server role. The role assumed depends on who sends the first message. In a communication between two UA, the first UA who sends the Q-HTTP BEGIN message for starting the handshake phase will assume the client role. If both send the message at the same time, then both will wait a random time to restart again. Otherwise, an UA may be configured to act only as server (e.g., content provider's side). Garcia Aranda Expires May 8, 2011 [Page 52] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 +------------------------------------------------+ | | | UA(Client) UA(Server) | | | | -------- Q-HTTP BEGIN -------------> | | <------- Q-HTTP BEGIN -------------- | | | | ------- Q-HTTP BEGIN --------------> | | <------ Q-HTTP 200 OK -------------- | | | | | +------------------------------------------------+ Figure 17 P2P roles. 5.2. Multiple Quality sessions in parallel A quality session is intended to be used for an application. It means that for using the application, the client MUST establish only one quality session against the server. Indeed, the relation between session-id and application is 1 to 1. If a user wants to participate in several independent quality sessions simultaneously against different servers (or against the same server) can execute different Q-HTTP clients to establish separately different quality sessions but it is not recommended, because: o The establishment of a new quality session may affect other running applications over other quality sessions. Thus, minimum quality level may not be achieved depending on individual requirements of each application. o If the negotiation phase is executed separately before running any application, the quality requirements could not be assured when the applications are running in parallel. For running different applications in parallel it is highly recommended to execute the negotiation phase of all of them simultaneously, in order to assure the quality constraints of all applications in parallel. To do that, a single User Agent software MUST be used, and this User Agent MUST be able to launch several quality session negotiation in parallel, synchronizing the beginning of each negotiation phase, and running again the negotiation phase of all applications in parallel until all of them succeed. Garcia Aranda Expires May 8, 2011 [Page 53] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 In order to repeat the execution of a negotiation phase that has been succeeded, both, client and server MUST allow using the READY method with a Stage header value already succeeded. 5.3. General client behavior An Q-HTTP Client has different behaviors. We will use letters X,Y,Z for designate each different behavior (follow the letter bullets in the figure below). X) When it sends messages over TCP (methods GET, QOS-ALERT and CANCEL) behaves strictly like a state machine which sends messages and wait for responses. Depending on the response type it enters in a new state. When sends UDP messages (methods PING and DATA), a Q-HTTP client is not strictly a state machine which sends messages and wait for receiving responses because: Y) At latency, jitter and packet loss measurement, the PING requests (over UDP) are sent periodically, not after receiving the response to the previous request. In addition, the client MUST answer the PING messages received from server, therefore assumes the role of a server. Z) At bandwidth and packet loss measurement stage, the client does not expect to receive responses when sends DATA requests (over UDP) to the server. In addition, it MUST receive and process all server messages in order to achieve the downlink measurement. In addition to this special behavior, the methods QOS-ALERT and CANCEL have successful responses which follow the same syntax rules of a request (instead of 2xx response code). However, these methods may have a conventional answer if an error is produced. Garcia Aranda Expires May 8, 2011 [Page 54] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 +-----------+------------------------+-----------+ | Handshake | Negotiation |Continuity | | Phase | Phase | Phase | | | | | | X ---------> Y --> X --> Z --> X ---> Y --> X | | | A | A | | A | | | | | | | | | | | | | | +-----+ +-----+ | +-----+ | | | | | +------------------------------------------------+ Figure 18 Phases & client behaviors. 5.3.1. Generating requests A valid Q-HTTP request formulated by a Client MUST, at a minimum, contain the following header fields: If no SDP is included: This is the case of PING and DATA messages. The header Session-id and Message-id are mandatory. If SDP is included: this is the case of GET, QOS-ALERT and CANCEL messages. Inside SDP is included Session-id, therefore the inclusion of session-id header is optional. 5.4. General server behavior If a Server does not understand a header field in a request (that is, the header field is not defined in this specification or in any supported extension), the server MUST ignore that header field and continue processing the message. The role of server is changed at negotiation and continuity phases, in which server MUST send packets to measure jitter, latency and bandwidth. Therefore, the different behaviors of server are (follow the letter bullets in the figure below): R) When the client sends messages over TCP (methods GET, QOS-ALERT and CANCEL) behaves strictly like a state machine which receives messages and sends responses. Garcia Aranda Expires May 8, 2011 [Page 55] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 When the client begins to send UDP messages (methods PING and DATA), a Q-HTTP server is not strictly a state machine which receives messages and sends responses because: S) At latency, jitter and packet loss measurement, the PING requests (over UDP) are sent periodically by the client but also by the server. In this case the server behaves as a server answering client requests but also behaves as a client, sending PING messages toward the client and receiving responses. T) At bandwidth and packet loss measurement, the server sends DATA requests (over UDP) to the client. In addition, MUST receive and process client messages in order to achieve the uplink measurement. In addition to this special behavior, the methods QOS-ALERT and CANCEL have successful responses which follow the same syntax rules of a request (instead of 2xx response code). However, these methods may have a conventional answer if an error is produced. +-----------+------------------------+-----------+ | Handshake | Negotiation |Continuity | | Phase | Phase | Phase | | | | | | R ---------> S --> R --> T --> R ---> S --> R | | | A | A | | A | | | | | | | | | | | | | | +-----+ +-----+ | +-----+ | | | | | +------------------------------------------------+ Figure 19 Phases & server behaviours. 6. Q-HTTP method definitions The Method token indicates the method to be performed on the resource identified by the Request-URI. The method is case-sensitive. Garcia Aranda Expires May 8, 2011 [Page 56] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 Method = "BEGIN" | "PING" | "DATA" | "GET" | "QOS-ALERT" | "CANCEL" | "READY" | extension-method extension-method = token The list of methods allowed by a resource can be specified in an "Allow" header field (RFC 2616 [1] section 14.7). The return code of the response always notifies the client when a method is currently allowed on a resource, since the set of allowed methods can change dynamically. Any server application SHOULD return the status code 405 (Method Not Allowed) if the method is known, but not allowed for the requested resource, and 501 (Not Implemented) if the method is unrecognized or not implemented by the server. 6.1. BEGIN The BEGIN method means request information from a resource identified by a q-http URI. The semantics of this method is the starting of a quality session. This method is only used in handshake phase to retrieve the SDP containing all quality parameters for the desired application to run. In the negotiation and continuity phases, this method is not used. when a BEGIN message is received by the server, any current quality session is cancelled and a new session should be created. The response to a Q-HTTP BEGIN request is not cacheable. 6.2. GET The GET method means retrieve information from a resource identified by a q-http URI. In the negotiation and continuity phases, this method is used to check if the server considers the quality good enough to execute the desired application. If the measured quality is not enough, the server will return a 412 error. The response to a Q-HTTP GET request is not cacheable. 6.3. READY The READY method is used to synchronize the starting time for sending of PING and DATA messages over UDP between client and servers. Garcia Aranda Expires May 8, 2011 [Page 57] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 In addition, the Stage header included in this method is mandatory and allow clients repeat a test, which is needed in scenarios with multiple quality sessions between one client and different servers. This message is only used in negotiation and continuity phases, and only just before making a measurement. Otherwise (out of this context), the server MUST ignore this method. 6.4. PING This message is used to measure the RTT and jitter of a session. The message MUST be sent only over UDP control port. If a server receives this message in other port it MUST ignore it. The fundamental difference between the PING and DATA requests is reflected in the different measurements achieved with them. PING is a short message, and MUST be answered in order to measure RTT, whereas DATA is a long message (1 Kbyte) and MUST NOT be answered. PING is a request method that can be originated by client but also by server. Client MUST answer the server PINGs, assuming a "server role" for these messages during measurement process. 6.5. DATA This message is used to measure the bandwidth and packet loss of a session. The message MUST be sent only over UDP control port. If a server receives this message in other port it MUST ignore it. The fundamental difference between the PING and DATA requests is reflected in the different measurements achieved with them. PING is a short message, and MUST be answered in order to measure RTT, whereas DATA is a long message (1 Kbyte) and MUST NOT be answered. DATA is a request method that can be originated by client but also by server. Both (client and server) MUST NOT answer DATA messages. 6.6. QOS-ALERT This is the message that Q-http generates when the measurements indicate that quality SLA is being violated. It is an informative message which indicates that the user's experience is being degraded and includes the details of the problem (bandwidth, jitter, packet loss measurements and the SLA). The QoS-ALERT message does not contain any detail on the actions to be taken, which depends on the agreements between all involved parties. Garcia Aranda Expires May 8, 2011 [Page 58] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 The message has no conventional answer, but a request format message is answered from the server when it receives a client QOS-ALERT. This method can be initiated by client only after a 412 error coming from server, and with enough information to build the QOS-ALERT message. If the header "Q-HTTP-policy-server" was included in the server response of the handshake phase, the QOS-ALERT message MUST be sent to the URI indicated in this header, otherwise the QOS-ALERT message MUST be sent to the server. With policy server, the QOS-ALERT message sent by client MUST contain the URIs of the server and the client to be contacted later by the policy server. Therefore the following headers MUST be included in the client request: "Q-HTTP-Resource-server" and "Q-HTTP-Resource- client". The response to a Q-HTTP QOS-ALERT request is not cacheable. 6.7. CANCEL Like QOS-ALERT, this message is used for communication with the network resources. The semantics in this case is the release of the special resources assigned to the session. In the same way as QOS-ALERT, CANCEL has the same type of response, with a request format. 7. Response codes All Q-HTTP response codes are used only in TCP control flows. Never in UDP message flows, which are used for measurements. 7.1. 100 trying This response indicates that the request has been received by the next-hop server (the policy server) and that some unspecified action is being taken on behalf of this request (for example, a database is being consulted). This response, like all other provisional responses, stops retransmissions of a QOS-ALERT by the client. 7.2. 200 OK The request has succeeded. Garcia Aranda Expires May 8, 2011 [Page 59] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 7.3. Redirection 3xx 3xx responses give information about the user's new location, or about alternative services that might be able to satisfy the request. The requesting client SHOULD retry the request at the new address(es) given by the Location header field. 7.4. Request Failure 4xx 4xx responses are definite failure responses from a particular server. The client SHOULD NOT retry the same request without modification (for example, adding appropriate headers or SDP values). However, the same request to a different server might be successful. 7.4.1. 400 Bad Request The request could not be understood due to malformed syntax. The Reason-Phrase SHOULD identify the syntax problem in more detail, for example, "Missing Message-id header field". 7.4.2. 404 Not Found The server has definitive information that the user does not exist at the domain specified in the Request-URI. This status is also returned if the domain in the Request-URI does not match any of the domains handled by the recipient of the request. 7.4.3. 405 Method Not Allowed The method specified in the Request-Line is understood, but not allowed for the address identified by the Request-URI. The response MUST include an Allow header field containing a list of valid methods for the indicated address. 7.4.4. 406 Not Acceptable The resource identified by the request is only able of generating response entities that have content characteristics not acceptable according to the Accept header field sent in the request. 7.4.5. 408 Request Timeout The server could not produce a response within a suitable amount of time, and the client MAY repeat the request without modifications at any later time Garcia Aranda Expires May 8, 2011 [Page 60] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 7.4.6. 412 A precondition has not been met The server is indicating that the SLA is being violated. 7.4.7. 413 Request Entity Too Large The server is refusing to process a request because the request entity-body is larger than the one that the server is willing or able to process. The server MAY close the connection to prevent the client from continuing the request. 7.4.8. 414 Request-URI Too Long The server is refusing to process the request because the Request-URI is longer than the one that the server accepts. 7.4.9. 415 Unsupported Media Type The server is refusing to process the request because the message body of the request is in a format not supported by the server for the requested method. The server MUST return a list of acceptable formats using the Accept, Accept-Encoding, or Accept-Language header field, depending on the specific problem with the content. 7.4.10. 416 Unsupported URI Scheme The server cannot process the request because the scheme of the URI in the Request-URI is unknown to the server. 7.5. Server Failure 5xx 5xx responses are failure responses given when a server itself is having trouble. 7.5.1. 500 Server Internal Error The server encountered an unexpected condition that prevented it from fulfilling the request. The client MAY display the specific error condition and MAY retry the request after several seconds. 7.5.2. 501 Not Implemented The server does not support the functionality required to fulfill the request. This is the appropriate response when a Server does not recognize the request method and it is not capable of supporting it for any user. Garcia Aranda Expires May 8, 2011 [Page 61] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 Note that a 405 (Method Not Allowed) is sent when the server recognizes the request method, but that method is not allowed or supported. 7.5.3. 503 Service Unavailable The server is temporarily unable to process the request due to a temporary overloading or maintenance of the server. The server MAY indicate when the client should retry the request in a Retry-After header field. If no Retry-After is given, the client MUST act as if it had received a 500 (Server Internal Error) response. A client receiving a 503 (Service Unavailable) SHOULD attempt to forward the request to an alternate server. It SHOULD NOT forward any other requests to that server for the duration specified in the Retry-After header field, if present. Servers MAY refuse the connection or drop the request instead of responding with 503 (Service Unavailable). 7.5.4. 504 Server Time-out The server did not receive a timely response from an external server it accessed in attempting to process the request. 7.5.5. 505 Version Not Supported The server does not support, or refuses to support, the Q-HTTP protocol version that was used in the request. The server is indicating that it is unable or unwilling to complete the request using the same major version as the client, other than with this error message. 7.5.6. 513 Message Too Large The server was unable to process the request since the message length exceeded its capabilities. 7.6. Global Failures 6xx 6xx responses indicate that a server has definitive information about a particular policy not satisfied for processing the request. 7.6.1. 600 session not exist The session-id is not valid Garcia Aranda Expires May 8, 2011 [Page 62] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 7.6.2. 601 quality level not allowed The QOS level requested is not allowed for the pair client/server 7.6.3. 603 Session not allowed The session is not allowed due to some policy (number of sessions allowed for the server is exceeded, or the time band of the QOS-ALERT is not allowed for the pair client/server, etc) 7.6.4. 604 authorization not allowed The policy server does not authorize the QOS-ALERT operation because any internal or external reason. 8. Implementation Recommendations 8.1. Default client constraints To provide a default configuration, it would be good that the client had a configurable set of Quality headers in the browser settings menu. Otherwise these quality headers will not be present in the first message. Different business models (out of scope of this proposal) may be achieved: depending on who pays for the quality session, the server can accept certain Client parameters sent in the first message, or force billing parameters on the server side. 8.2. Bandwidth measurements In programming languages or Operating Systems with timers or limited clock limited resolution, it is recommended to use an approach based on several intervals to send messages of 1KB, in order to reach the required bandwidth consumption using a rate closest as possible to a constant rate. For example, if the resolution is 1 millisecond, and the bandwidth to reach is 11Mbps, a good approach consists of sending: 1 message of 1KB each 1 millisecond + 1 message of 1KB each 3 milliseconds + 1 message of 1KB each 23 milliseconds Garcia Aranda Expires May 8, 2011 [Page 63] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 The number of intervals depends on required bandwidth and accuracy that the programmer wants to achieve. 8.3. Packet loss measurement resolution Depending on application nature and network conditions, a packet loss resolution less than 1% may be needed. In such case, there is no limit to the number of samples used for this calculation. A tradeoff between time and resolution should be reached in each case. For example, in order to have a resolution of 1/10000, the last 10000 samples should be considered in the packetloss measured value. The problem of this approach is the reliability of old samples. If the interval used between PING messages is 50ms, then to have a resolution of 1/1000 it takes 50 seconds and a resolution of 1/10000 takes 500 seconds (more than 8 minutes). The reliability of a packet loss calculation based on a sliding window of 8 minutes depends on how fast network conditions evolve. 8.4. qos-level dictionary There is no precise meaning for each level at all, but only the principle that, in general, a higher level should correspond to a better quality. 8.5. Measurements and reactions Q-HTTP can be used as a mechanism for measure and trigger actions (i.e. lowering video bit-rate) in real-time in order to reach the application constraints, addressing measured possible network degradation. The trigger is based on message QOS-ALERT, which is always forced by the server response 412 error. A server can avoid these Q_OS-REQUEST messages sending 200 OK when a GET message is received from server, independently whether the constraints are met or not. 8.6. Scenarios Q-HTTP could be used in two scenarios: o client to ACP (Application content provider ) o client to client. Garcia Aranda Expires May 8, 2011 [Page 64] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 8.6.1. Client to ACP In this scenario, the policy server is optional. If it exists, the QOS-ALERT messages MUST be sent to this policy server which acts as a proxy for this type of messages and validates them (plus any other actions out of scope of this document). In order to avoid useless load on the server, the policy server could receive the BEGIN messages of handshake phase. For this purpose, the policy server MUST know the URI of the Q-HTTP servers. In this scenario a client could send the BEGIN to the policy server, with an additional parameter in the URI requested, which identifies the server, like: Httpq://www.policy.com/listofservers?id=xtiwn28821ho4 Then the policy validates the request and forward the BEGIN to the Q- HTTP server, adding the Q-HTTP-Resource-server to the response for the client in the 200 OK response. +------------------------------------------------+ | | | Client policy Server | | | | --- BEGIN ---------> | | <-- 100 trying ----- | | | | --- BEGIN ----------> | | <--- 200 OK ---------- | | <--- 200 OK----- --- | | | +------------------------------------------------+ Figure 20 Policy server. In this scenario the client MUST send further messages directly to the server without passing through policy server. 8.6.2. Client to client In order to solve the client to client scenario, a Q-HTTP register function MUST be implemented . This allows clients contact each other for sending the BEGIN message. In this scenario, the policy server Garcia Aranda Expires May 8, 2011 [Page 65] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 MUST complete the Q-HTTP-Resource server with the public IP address of the peer which assumes the server role. The register function is out of scope of this protocol version, because different HTTP mechanisms can be used and Q-HTTP MUST NOT force any. 9. Security Considerations Different types of attacks can be avoided: o Spoofing of server IP address can be avoided using the digital signature mechanism. The network can validate easily this digital signature using the public key of the server certificate. o The client could try to send ALERT messages constantly, trying to enter in the negotiation phase continuously. In this case, the server MUST answer a message "CANCEL", in order to release the all levels reached and return to plain access without enhanced quality. This protocol could be supported over IPSec to increase privacy, although it is out of scope of this proposal. 10. IANA Considerations A specific port for Q-HTTP TCP control flow mechanism could be assigned. It could simplify the network implementation. Other possibility is to use any other port (like 80, HTTP). In this case the network could use the protocol designator "Q-HTTP" as the mark for distinguish and treat the packets. Q-HTTP uses SDP as a container for session information, in which quality attributes have been added as extended "session-level" attributes. These set of new attributes should be registered (in order to avoid the prefix "X-"). In this document, this set of attributes has been presented as registered attributes. This is the list of attribute field names to register: Garcia Aranda Expires May 8, 2011 [Page 66] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 Attribute name : qos-level Type of attribute: session level subject to the charset attribute: NO explanation of purpose: define the current qos profile in uplink and downlink for the communication between client and server. The exact meaning of each level is implementation dependant but in general, a higher qos-level value corresponds to a better quality network profile. Appropriate attribute values: [0..9] "/" [0..9] Attribute name : latency Type of attribute: session level subject to the charset attribute: NO explanation of purpose: define the latency constraints in milliseconds in uplink and downlink for the communication between client and server. Appropriate attribute values: [0..9999] "/" [0..9999] If there is no constraint in some direction (uplink, downlink or both) the value can be empty in that direction Attribute name : jitter Type of attribute: session level subject to the charset attribute: NO explanation of purpose: define the jitter constraints in milliseconds in uplink and downlink for the communication between client and server. Appropriate attribute values: [0..9999] "/" [0..9999] Attribute name : bandwidth Type of attribute: session level subject to the charset attribute: NO explanation of purpose: define the bandwidth constraints in kbps in uplink and downlink for the communication between client and server. Appropriate attribute values: [0..99999] "/" [0..99999] Attribute name : packetloss Type of attribute: session level subject to the charset attribute: NO explanation of purpose: define the packet loss tolerance constraints in 100% in uplink and downlink for the communication between client and server. Appropriate attribute values: [0..99] "/" [0..99] Attribute name : flow Type of attribute: session level subject to the charset attribute: NO Explanation of purpose: define a flow between a client and a server. The flow involves purpose (data or control), direction (uplink or downlink) protocol (UDP or TCP) and port or range or ports Garcia Aranda Expires May 8, 2011 [Page 67] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 attribute values: <"control"|"data"> <"uplink"|"downlink"> <"UDP"|"TCP"> <0..65535>[ "-" [0..65535]] Attribute name : measurement Type of attribute: session level subject to the charset attribute: NO Explanation of purpose: define the procedure to measure the quality and the different values for each measurement Attribute values: "procedure/" | "latency "[0..9999] "/" [0..9999] | "jitter "[0..9999] "/" [0..9999] | "bandwidth "[0..99999] "/" [0..99999] | "packetloss "[0..99] "/" [0..99] If the attribute value is "procedure", the rest of the line MUST contain the name of the procedure and optional parameters, separated by ",". In the case of procedure "default", the valid values are: a=measurement:procedure default,[0..999]"/" [0..999] "," [0..999] "/" [0..999] "," [0..9999] "," [0|1] where: o The first parameter is the interval of time (in milliseconds) between PING messages in the negotiation phase. Forward (client to server) and reverse (server to client) values separated by "/". o The second parameter is the interval of time (in milliseconds) between PING messages in the continuity phase. Forward (client to server) and reverse (server to client) values separated by "/". o The third parameter is the time used to measure bandwidth during negotiation phase. In case of not present, a default value of 5000 ms will be assumed. o The fourth parameter indicates the mode for continuity phase (0 means "normal" and 1 means "sliding window"). In case of not be present, normal mode (default value of 0) will be assumed. Garcia Aranda Expires May 8, 2011 [Page 68] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 Other procedure names are allowed, but at least "default" procedure implementation is mandatory in client and servers. 11. Conclusions Q-http defines three phases with different purposes, and inside these phases the negotiated measurement procedure is used. Different measurement procedures can be used (even RTCP itself) inside Q-HTTP. Basically, Q-HTTP only defines how to transport SLA information and measurement results as well as providing some mechanisms for alerting. Q-http does not ask for resources. Q-HTTP only alerts if one (or some) of SLA quality parameters are being violated. Depends on server (Application content provider) to do something with this information and return it back to a SLA-compliant state. Garcia Aranda Expires May 8, 2011 [Page 69] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 12. References 12.1. Normative References [1] Fielding, R., Gettys, J., Mogul, J., Frystyk, H.,Masinter, L., Leach, P. and T. Berners-Lee, "Hypertext Transfer Protocol -- HTTP/1.1" RFC 2616, June 1999. [2] Handley, M. and V. Jacobson, "SDP: Session Description Protocol", RFC 4566, July 2006. [3] Bradner, S., "Key words for use in RFCs to Indicate RequirementLevels", BCP 14, RFC 2119, March 1997. [4] Berners-Lee, T., Fielding, R. and L. Masinter, "Uniform Resource Identifiers (URI): Generic Syntax", RFC 3986, January 2005. [5] Rosenberg, J. and H. Schulzrinne, "An Offer/Answer Model with SDP", RFC 3264, June 2002. [6] Rivest, R., "The MD5 Message-Digest Algorithm", RFC 1321, April 1992. [7] Johnsson, J., B. Kaliski, "Public-Key Cryptography Standards (PCS) #1: RSA Cryptography Specifications version 2.1", RFC 3447, February 2003. [8] Postel, J., "DoD Standard Transmission Control Protocol", RFC 761, January 1980. [9] Postel, J., "User Datagram Protocol", STD 6, RFC 768, August 1980. [10] Schulzrinne, H., Casner, S., Frederick, R., Jacobson, V. "RTP: A Transport Protocol for Real-Time Applications", RFC 3550, july 2003. [11] Yergeau, F., "UTF-8, a transformation format of ISO 10646", RFC 3629, November 2003. [12] Resnick, P., "Internet Message Format", RFC 5322, October 2008 Garcia Aranda Expires May 8, 2011 [Page 70] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 12.2. Informative References [13] Rosenberg, J., Schulzrinne, H., Camarillo, G., Johnston, A. Peterson, J., Sparks, R., Handley, M. and Schooler, E. , "SIP: Session Initiation Protocol", RFC 3261, June 2002. [14] Mathis, M., Semke, J., Mahdavi, J., Ott, T., "The Macroscopic Behavior of the TCP Congestion Avoidance Algorithm", Computer Communications Review, 27(3), July 1997. [15] Floyd, S., "HighSpeed TCP for a Large Congestion Windows", RFC 3649, December 2003. [16] Rhee, I., Xu, L., Ha, S., "CUBIC for Fast Long-Distance Networks", Internet-draft draft-rhee-tcpm-cubic-02, February 2009. [17] Sridharan, M., Tan, K., Bansal, D., Thaler, D., "Compound TCP: A New TCP Congestion Control for High-Speed and Long Distance Networks", Internet-draft draft-sridharan-tcpm-ctcp-02, November, 2008. Garcia Aranda Expires May 8, 2011 [Page 71] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 13. Acknowledgments Many people have made comments and suggestions contributing to this document. In particular, we would like to thank: Sonia Herranz Pablo, Clara Cubillo Pastor, Francisco Duran Pina, Ignacio Moreno Lopez, Michael Scharf and Jesus Soto Viso. Garcia Aranda Expires May 8, 2011 [Page 72] Internet-Draft The Quality Hypertext Transfer Protocol November 2010 14. Authors' Addresses Jose Javier Garcia Aranda Alcatel-Lucent C/Maria Tubau 9 28050 Madrid Spain Phone: +34 91 330 4348 Email: Jose_Javier.Garcia_Aranda@alcatel-lucent.com Jacobo Perez Lajo Alcatel-Lucent C/Maria Tubau 9 28050 Madrid Spain Phone: +34 91 330 4165 Email: Jose_Javier.Garcia_Aranda@alcatel-lucent.com Luis Miguel Diaz Vizcaino Alcatel-Lucent C/Maria Tubau 9 28050 Madrid Spain Phone: +34 91 330 4871 Email: Luismi.Diaz@alcatel-lucent.com Garcia Aranda Expires May 8, 2011 [Page 73]