| Network Working Group | M. Boucadair |
| Internet-Draft | D. Binet |
| Intended status: Informational | C. Jacquenet |
| Expires: September 3, 2015 | France Telecom |
| L. Contreras | |
| Telefonica I+D | |
| March 2, 2015 |
On the Need for Transport Protocol Profiles & Investigating New Evolution Tracks
draft-boucadair-transport-protocols-00
The world of Internet transport protocols is changing, after decades of TCP and UDP operation. Several proposals have been submitted for the past years (and counting) to introduce other transport protocols that aim at reducing the web latency of that of TCP or avoiding the burden of the various middle-boxes (NATs, firewalls, for one) encountered along the communication path. Such initiatives, although not new, are motivated by the complexity of some (non-transparent) networking functions.
This document advocates for the definition of transport profiles and the need to document recommendations for middleboxes, including Performance Enhancement Proxies (PEPs) behaviors. A collaboration among the involved players (service providers, vendors) is required to soften the current complications encountered in the Internet at large.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.
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."
This Internet-Draft will expire on September 3, 2015.
Copyright (c) 2015 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. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
The world of Internet transport protocols is changing, after decades of TCP and UDP operation. Several proposals have been submitted for the past years (and counting) to introduce other transport protocols or additional features to existing protocols that aim at reducing the web latency of that of TCP or avoiding the burden of the various middle-boxes (NATs, firewalls, for one) encountered along the communication path. Such initiatives, although not new, are motivated by the complexity of some (non-transparent) networking functions. Further collateral effects (including a thorough identification of various network hindrances) are discussed in this document together with potential contributions from network operators to overcome some of the encountered issues.
Advanced service functions (e.g., Performance Enhancement Proxies ([RFC3135]), NATs, firewalls, etc.) are now required to achieve various objectives such as IP address sharing, firewalling, to avoid covert channels, to detect and protect against ever increasing DDoS attacks, etc. Removing those functions is not an option because they are used to address constraints that are often typical of the current yet protean Internet situation (global IPv4 address depletion comes to mind, but also the plethora of services with different QoS/security/robustness requirements, etc.), and this is even exacerbated by environment-specific designs (e.g., the nature and the number of service functions that need to be invoked at the Gi interface of a mobile infrastructure). Moreover, these sophisticated service functions are located in the network but also in service platforms, or intermediate entities (e.g., CDNs). This situation clearly complicates diagnostic procedures whenever service degradation is experienced, given That the responsibility is often shared among various players.
Also, there are performance issues that are specific to some wireless networks [I-D.manyfolks-gaia-community-networks].
An important effort was conducted by the IETF (e.g., BEHAVE, PCP, Performance Implications of Link Characteristics (pilc)), but we believe further work is still required to mitigate/soften some of the pending issues.
Note,
This document advocates for the definition of transport profiles and the documentation of recommendations for middleboxes, including Performance Enhancement Proxy (PEPs) behaviors. A collaboration among the usual players is required to soften the current complications encountered in the Internet at large.
Transport services refer to the set of features that are offered by protocols used to multiplex connections over IP. Examples of transport services include - but are not limited to- ordering delivery, reliable delivery, congestion control, or full or partial integrity protection.
A transport protocol can be abstracted as an implementation which exposes a set of transport services. For example, TCP (Transmission Control Protocol, [RFC0793]), which is the universally deployed and implemented transport protocol, offers reliable and ordered delivery, flow and congestion control, as well as primitives to manage a connection. Unlike TCP, UDP (User Datagram Protocol, [RFC0768]) is a connectionless protocol that supports protection against data corruption using a checksum field.
Given the hurdles induced by advanced network-located service functions, “Make your own protocol” is not even an option.
“Encapsulate over your favorite existing protocol”, if transported over TCP, has more chances to experience less session failures.
The strategy that consists in “extending your favorite widely deployed transport protocol” is more viable from a deployment perspective.
Plethora of transport protocols have been proposed by the Internet community to accommodate requirements raised by emerging applications. Overall, these applications are either requiring more transport services than what is actually offered by TCP and UDP, or less transport services.
For example, SCTP (Stream Control Transmission Protocol, [RFC4960]) was specified to accommodate applications which need more transport services than what can be offered by TCP (e.g., preserve (application) data boundaries, support of out-of-order delivery, built-in support of multiple streams).
DCCP (Datagram Congestion Control Protocol, [RFC4340]) is another protocol that was promoted to accommodate requirements from applications which need more transport services than what is offered by UDP (e.g., congestion control), but without suffering from the constraints of a connection-oriented protocol like TCP (e.g., reliable delivery mechanisms).
UDP-lite ([RFC3828]) is a light version of UDP that was designed for applications that need less features than what is offered by UDP (e.g., partial data corruption detection), whereas DTLS (Datagram Transport Layer Security, [RFC6347]) and TLS ([RFC5246]) were specified for applications requiring encryption capabilities at the transport layer.
Other candidate transport protocols are currently investigated to reduce the delay required to invoke a resource located in the Internet. Typically, this consists in retrieving some contents by minimizing the delay induced by TCP or SCTP handshakes required for establishing a connection. Yet, such approaches can take advantage of the transport services provided by connection-oriented protocols.
It is worth mentioning that reducing the delay to access a requested resource is not only the responsibility of transport protocols, but also depends on various other services such as DNS and access service functions. The whole chain should be optimized! Reduce the delay when invoking a service objective should be moderated with other considerations such as policy enforcement at the server side (including rate-limit and actions taken to protect against DDoS attacks).
Despite the effort made by the Internet community to specify new transport protocols or propose improvements of existing ones (mainly TCP), the deployment reality is that TCP remains hegemonic. Even worse, only connections destined to some TCP port numbers are allowed in some networks.
Recent studies (e.g., [Traffic]) revealed that TCP accounts for 84.35% of the total amount of packets forwarded over the Internet and 92% of the bytes. DCCP and SCTP were not found in those studies.
The main reasons that explain the poor adoption of new transport-related features at the scale of the Internet are:
Typical examples of service functions include: traditional NAT (Network Address Translation, [RFC3022]), CGN (Carrier Grade NAT; including IPv4-IPv4 CGN ([RFC6888]), DS-Lite AFTR ([RFC6333]) or NAT64 ([RFC6146])), firewall, application proxies, Performance Enhancement Proxies (PEP, [RFC3135]), traffic uniformizers, etc.
Transport-related solutions that assume that the remedy to the problem formulated above would be to withdraw all flow-aware service functions are not realistic. The presence of advanced service functions must be considered by solution designers as the rule rather than the exception.
Obviously, this does not mean that network providers should not question the pertinence to maintain some of these service functions active. Even if a rationalization effort is required in this area (still this is deployment-specific), solution designers should propose solutions that are robust in the presence of these functions.
For example, extensions have been proposed to enhance user's quality of experience when TCP is in use such as: TCP Fast Open ([RFC7413]), Proportional Rate Reduction ([RFC6937]), increase the initial window size ([RFC6928]), TCP Extensions for high performance ([RFC7323]) , unordered TCP/TLS, etc. More can be found in [RFC7414].
These extensions may have merits when taken individually, but the question is whether those merits are still valid when co-existing with other features. In addition, these merits are a function of the deployment context (for example in fixed or mobile networks).
Implementing small changes at large is here to stay. Moreover, changing a transport protocol stack may is subject to the amplification principle (See Section 2.2.1 of [RFC3439]) since changes may not only have local impacts but may also impact the stability of a network (e.g., MPTCP hosts are more aggressive than TCP hosts). Assessing the impact of these extensions on legacy hosts is critical.
According to [Traffic],
[Traffic] also showed that disordering is deployment-specific (because it was observed only in some networks); means that lead to such behavior should be disabled in those networks. This suggests reliable means to minimize such risks.
This data shows that several of TCP advances (e.g., WS) are not massively deployed or not deployed at all (e.g., ECN). More effort is required to evangelize recent TCP advances and their motivations.
Fortunately, there is still an opportunity for network providers to contribute to the improvement of transport services. A technical strategy that would focus on the root causes to properly derive associated recommendations should be encouraged.
Every (new) transport protocol will come with its own problems and perfectible features. Too many transport protocols are really painful for all actors, including for network operators (think about the configuration of class of services, fairness access and usage of network resources, and other traffic management services).
Leveraging skills and experience of TCP design as well as operation is a first major step for network providers. For example, in order to reduce latency for TCP-based applications, the following technical tracks can be investigated:
Network Providers should be able to keep on delivering differentiated services as a competitive business advantage, while mastering the complexity of the applications and enhancing customer's quality of experience. This can be achieved by exposing and communicating reachability information (i.e., routes to access desired contents) that will foster session establishment. This can be achieved using dedicated interfaces that can be used by applications.
Reduce complexity at the application level, strengthen the collaboration between the applications and the network layer via clear interfaces should also be encouraged, but this may be subject to agreements. Administrative-related considerations are out of scope of this document.
From a network provider perspective, the following risks need to be taken into account when designing solution(s) that would enhance current transport services:
Some recommendations to improve transport services have been documented for quite some time (e.g., [RFC4787], [RFC5382]).
Such recommendations are related to the design and the operation of services in the presence of flow-aware devices (in particular, NATs). A few examples: the use of endpoint-independent NAT mapping (EIM) and filtering (EIF) behaviors, IP address pooling behavior of "Paired" to not break protocols such as RTP/RTCP, the selection of long mapping lifetime values to avoid breaking some applications, the preservation of port parity for RTP/RTCP-based applications (like VoIP), the preservation of port contiguity for some applications, the use of port randomness to avoid session hijacking, the ability to discover the external IP address/port/lifetime ([RFC6887]) so that applications with referral behave with no degradation, the analysis of the use of the HOST_ID ([RFC6967]) to soften issues induced by address sharing at large ([RFC6269]), etc.
An effort to clarify some of the behave requirements is ongoing ([I-D.ietf-tsvwg-behave-requirements-update]).
Also, the Performance Implications of Link Characteristics (pilc) WG conducted an important effort which led to [RFC3135][RFC3150][RFC3155][RFC3366][RFC3449][RFC3481][RFC3819].
The following candidate actions are proposed (non-exhaustive list):
This document makes no request of IANA.
Add some text about privacy and security.
TBC.