INTERNET DRAFT draft-esaki-co-cl-ip-forw-atm-00.txt H. Esaki K. Nagami Toshiba Corporation M. Ohta Tokyo Institute of Technology October 14, 1994 Connection Oriented and Connectionless IP Forwarding Over ATM Networks Status of this Memo This document is an Internet-Draft. 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. Internet-Drafts may be updated, replaced, or obsoleted by other documents at any time. It is not appropriate to use Internet- Drafts as reference material or to cite them other than as a ``working draft'' or ``work in progress.'' To learn the current status of any Internet-Draft, please check the 1id-abstracts.txt listing contained in the Internet-Drafts Shadow Directories on ds.internic.net, nic.nordu.net, ftp.nisc.sri.com, or munnari.oz.au. Abstract The framework of high throughput with small latent IP packet delivery architecture using ATM technology in large scaled heterogeneous internets keeping the current commubication paradigm is studied. This memo discusses on how to carry TCP/IP over ATM, or, in general, how to carry connectionless and/or connection oriented network layer protocol over connection oriented QoS-ed datalink layer. Low latency property of ATM is fully extracted both for connectionless and connection oriented communication over TCP/IP without changing the current architecture of TCP/IP network. Regarding connection oriented IP packet forwarding, source host, having the mapping functionality between flow-ID (e.g. in the SIPP16 header) and VCI/VPI value, forwards connection oriented IP packets cell-by-cell. Connectionless IP packets can be also transferred cell-by-cell through the dedicated (concatenated) ATM-VCCs which bypass IP processing at the intermediate routers, when the cache for re-directed ATM-VCCs is hit. Otherwise, they will be forwarded to the adjacent router (hop-by-hop IP forwarding). Also, the active connectionless IP packet flow can use a Esaki, Nagami & Ohta Expires on April. 20, 1995 [Page 1] INTERNET DRAFT CO&CL IP Forwarding Over ATM Network October1994 dedicated ATM-VCC as a IP forwarding cache. Connectionless IP packet is forwarded cell-by-cell, instead of packet-by-packet, as far as the connectionless IP packet flow is in active state. 1. Introduction --------------- ATM is recognized as one of the platforms to provide high speed QOS (Quality of Service) data-link layer service. Here, the QOS data-link layer service can be end-to-end, rather than link-by-link, in ATM networks. But, it should be still recognized as a data-link layer service, rather than network layer service. ATM network provides both connection oriented service and connectionless network leyer service (e.g.current IP forwarding) topping on connection oriented VCCs, which could have a certain QOS. Physical or logical data-link network segments, including ATM-LANs, will be interconnected through routers (network layer entity), even when the ATM becomes major data-link platform. This draft proposes high speed and small latent IP packet delivery architecture over the Internet, using ATM technology. Regarding connectionless service, we can provide the small latent IP packet delivery using ATM bypass route, not only for IPv6 network but also for IPv4 network. And, connectionless IP packet can be forwarded cell-by-cell using the dedicated (concatenated) ATM-VCC, as far as the connectionless IP packet flow is in active state. Regarding connection oriented (network layer) service, we can provide high throughput small latent IP packet delivery using end-station's mapping functionality between flow-ID in the IP packet header and VCI/VPI value. Also, when router has a mapping capability between the network layer QOS parameter (flow-spec) and the ATM's QOS parameter, end-to-end network layer QOS will be provided. The proposed architecture can be applied to all of ATM network architecture discussed in IETF and in ATM Forum [LANE][CLIP][MBMA] [NHRP]. The proposed architecture can improve the performance of IP forwarding beyond IP subnet and can include short cut routing discussed in [NHRP]. Here, though this draft only discusses regarding ATM technology, the technique discussed in this draft can be generally applied to other communication platforms that are based on connection oriented technology (e.g. Frame Relay). 2. Internet Architecture including ATM-LANs ------------------------------------------- 2.1. Network Architecture in ATM Forum and IETF Models developed at LE-WG (LAN Emulation Working Group) of ATM Esaki, Nagami & Ohta Expires on April. 20, 1995 [Page 2] INTERNET DRAFT CO&CL IP Forwarding Over ATM Network October1994 Forum and at ATM-WG and ROLC(Routing Over Large Cloud)-WG of IETF are discussed [CLIP][NBMA]. Basically following three architecture models are discussed. Let say them as CLIP-model, NBMA-model and COIP-model, in this draft. We did not discuss LAN Emulation Model of ATM Forum, because the essential issue addressed here in LAN Emulation Model of ATM Forum is the same as the issue in CLIP-model. 2.1.0. IP-model To operate a network with a lot of hosts and some redundant paths, the network should divided into several data-link network segments, which are interconnected by routers. It would be recommended that each data-link network segment should accommodates only one IP subnet (network layer segment). This is the current network architecture in the Internet. And, both from the distributed application's point of view and from the network management's point of view, the data-link network segment (including ATM-LAN) should have broadcast capability. This is because broadcast is the only way to communicate with other end-stations (or routers) without prior knowledge of their addresses. 2.1.1. CLIP-model [CLIP] Generally, (logical) IP subnets will overlay topping on a single ATM network segment, which may be large scaled network. This means that several end-stations which generally belong to different IP subnets will exist on a single (ATM-ed) data-link segment. IP subnets are interconnected by the routers, which are completely the same as the current routers. The router in [CLIP] has only IP forwarding capability, but it does not have cell relaying capability. This means that ATM-VCC shall be always terminated at router. Therefore, it is impossible to provide end-to-end QOS-ed virtual connection using CLIP-model, without some protocols discussed in IETF, e.g.[RSVP]. Here, even when RSVP is applied in CLIP model, datagram transferring is still performed by IP packet level and it causes both delay and HOL (Head Of Line) blocking at router. Also, since the IP packet must be always reassembled at the router (i.e. packet-by-packet forwarding), every IP packet experiences IP packet reassembling delay. 2.1.2. NBMA-model [NBMA][NHRP] Logical IP subnets will overlay on a single ATM data-link network segment. NBMA-subnet is defined as that the network domain where end-to-end ATM-VCC can be established. Within the NBMA-subnet, each end-station can establish (seamless) ATM-VCC toward any end-host that is located at other IP subnet, without passing through a router, i.e. short-cut routing. However, when the IP packet should be transferred to out-side the NBMA-subnet, IP packet will be transferred to the boarder router between two NBMA-subnet. This means that, even if the neighbor NBMA-subnet Esaki, Nagami & Ohta Expires on April. 20, 1995 [Page 3] INTERNET DRAFT CO&CL IP Forwarding Over ATM Network October1994 is also ATM network, it is impossible to establish end-to-end ATM-VCC using IP address. Then, we can not obtain high performed IP packet forwarding beyond NBMA subnet. Here, the segmentation of the large scaled network into multiple NBMA subnets may be realized due to security concern. Therefore, the routers locating at the edge of NBMA subnets may perform security functionaloties, e.g. fire-wall function. 2.1.3. COIP-model [Ohta] COIP-model is just same as IP-model for all the connectionless and most of the connection oriented IP flows. But, some connection oriented traffic may be mapped to multiple datalink layer ATM VCCs. On routers, ATM-VCCs are conceptually terminated at router but actually relayed cell-by-cell. As the model allows moderately small datalink layer without losing direct connectivity of ATM, it is possible to introduce subnet-wide broadcast capability. 2.2. Internet Architecture including ATM Platforms The network segment (autonomous system) using ATM technology will increase in the future. However, ATM-LANs are interconnected each other vis routers, as well as other network platform (e.g.high speed Ethernet, FFOL or Fibre Channel). And, the old network platform must be smoothly replaced by ATM technology or by the other high speed platforms. Also, from the network operation point of view, the each network segment would be rationally small, and be operated independently. Since transmission of cells over WAN may be costly, outgoing (or incoming) cell VCCs would be aggregated into fewer VCCs. In a WAN environment, the user may want to open very few VCs and cell flows from a lot of end-stations would be connected to few VCCs to cross the WAN. Also, the signaling protocol would be different between the WAN environment and LAN environment. ATM-LAN is composed of switch nodes, and the interface between the adjacent network segment is UNI or NNI. Data-link segment may be identical to IP subnet. Or, the IP subnets may overlay over large cloud data-link network. But, IP subnets are interconnected by routers. And, data-link network segment may have multiple switching nodes (cell-relaying entities), or may have only a single switching node in it. The latter case is that every switching node has both cell-relaying functionality and IP forwarding functionality so as to concatenate individual ATM-VCCs which are provided by ATM-LANs, while each switch node has its own IP subnet. The key functionalities of the proposed router, that interconnects ATM networks (below represented as ATM router) are as follows. (1) Cell Relaying Cells are relayed as ATM layer's function. Since the assignment of VCI/VPI value is performed independently at individual ATM-LAN, router has the VCI/VPI translation functionality. Since IP processing can be bypassed by Esaki, Nagami & Ohta Expires on April. 20, 1995 [Page 4] INTERNET DRAFT CO&CL IP Forwarding Over ATM Network October1994 ATM layer cell relaying at router, the transmission latency to pass through the router can be sufficiently small, which is the same as an ATM switch node. (2) End-to-end connection oriented data transfer channel set-up During a set-up procedure of end-to-end connection oriented (or sometimes said as resource reservation oriented) data transfer channel, e.g.RSVP with SIPP, routers co-operate with the connection (ATM-VCC as data-link connection) establishment servers' in the ATM-LANs. And, also, we can achieve higher throughput than the conventional hop-by-hop IP forwarding by the router can. For example, the VCI/VPI translation table that concatenates ATM-VCCs provided by indvidual ATM-LANs will be established during the channel set-up procedure. (3) VCC multiplexing/de-multiplexing When communication cost over the WAN is costly, cell flows crossing WAN would be aggregated at the router. And, aggregated packet flows into few ATM-VCC would be de-multiplexed at the router. (4) Security Functionality (if necessary) Since every cell (or could be say packet) passes through the router, the segmentation of subnet can be fulfilled as well as the current Internet architecture. For instance, the proposed router can equip security functionalities, e.g. packet filtering in fire-wall function. The proposed router can equip security functionalities, that is, unauthorized packets are filered just as usual firewall and unauthorized reservation request packets for cell relaying are rejected. Routers are network entity which interconnect several IP subnets. Routers exchange routing information and maintain routing information tables to forward the received packet. Logically, each packet is relayed to the optimal interface by looking up the routing tables. But, actually, routers don't have to consult with routing table everytime they receive a packet. Implementations are free to have internal cache or bypass to minimize the packet routing delay of complex table looking up. Likewise, the proposed routers don't have to consult routing table everytime they receive a cell. Implementations are free to have internal bypass to remove the packet reconstruction delay and packet routing delay of complex table looking up. 3. Network Reference Model -------------------------- The following four are the network reference models discussed in this draft. In the following referenced models, every router in each model is the router that this draft proposes. Esaki, Nagami & Ohta Expires on April. 20, 1995 [Page 5] INTERNET DRAFT CO&CL IP Forwarding Over ATM Network October1994 3.1. CLIP model Large cloud network accommodates multiple logical IP subnets (LISs). In the figure, two sets of router and end-host belonging to the different LISs are shown. IP packet from host A.1 toward host B.1 passes through Router A.0 and Router B.0. When the IP packet flow from host A.1 to host B.1 is either active flow or connection oriented IP transport flow, the IP packet will be forwarded without IP forwarding process in the router. Also, IP packet can be forwarded through a direct ATM-VCC from Router A.0 to Router B.0. Here, IP packet may pass through the intermediate routers, that forwards IP packet cell-by-cell (not packet-by-packet through IP forwarding process), to reach at Router B.0. < Router A.0 > +----------------+ | +------------+ | | | IP forward | | | +------+-----+ | | | | | +------+-----+ | | | cell relay | | +----------+ | +------------+ | | Host_A.1 | +----------------+ +----------+ | | | | +------------------------------+ | | | | | Large Cloud Network | |(ATM-VCC switching capability)| | | | | +------------------------------+ | | | | +----------------+ +----------+ | +------------+ | | Host_B.1 | | | cell relay | | +----------+ | +------+-----+ | | | | | +------+-----+ | | | IP forward | | | +------+-----+ | +----------------+ < Router B.0 > (FIG.1 CLIP model) ^^^^^^^^^^^^^^^^^ Esaki, Nagami & Ohta Expires on April. 20, 1995 [Page 6] INTERNET DRAFT CO&CL IP Forwarding Over ATM Network October1994 3.2. NBMA model NBMA subnet accommodates multiple logical IP subnets (LISs). In the figure, two end-hosts belonging to the different LIS, but belonging to the same NBMA subnet are shown. In order to transfer the IP packet (both connection oriented IP packet and connectionless IP packet) from host A.x.1 to host A.y.1, host A.x.1 resolve the ATM address of host A.y.1 or the ATM address of some entity (e.g. router similar in CLIP model) relaying the IP packets to host A.y.1 using NHRP server. Then, IP packet from host A.x.1 to host A.y.1 could pass through so called short-cut route. Short-cut route may bypass the intermediate router(s), physically. IP packet from host A.x.1 toward host B.z.1 passes through Router, which is located at the edge point of NBMA subnets. When the IP packet flow from host A.x.1 to host B.z.1 is either active flow or connection oriented IP transport flow, the IP packet will be forwarded without IP forwarding process in the router, as well as in the NBMA subnet. +------------+ | Host_A.x.1 | +------------+ | +-------------------+ | | +-------------+ | +----| NHRP Server | | NBMA subnet A.0 | +-------------+ | | +------------+ | +----| Host_A.y.1 | | | +------------+ +-------------------+ | +-------------------------+ | +-------+ +---------+ | | | cell | | IP | | <----------- Router | | relay +---+ forward | | | +-------+ +---------+ | +-------------------------+ | +-------------------+ | | +------------+ | +----| Host_B.z.1 | | NBMA subnet B.0 | +------------+ | | +-------------+ | +----| NHRP Server | | | +-------------+ +-------------------+ (FIG.2 NMBA model) ^^^^^^^^^^^^^^^^^ Esaki, Nagami & Ohta Expires on April. 20, 1995 [Page 7] INTERNET DRAFT CO&CL IP Forwarding Over ATM Network October1994 3.3. COIP model Data-link network segment is identical to IP subnet. Therefore, IP packet beyond IP subnet must always pass router which is located at the edge of data-link network segments. In order to transfer the IP packet (both connection oriented IP packet and connectionless IP packet) from host A.1 to host B.1, host A.1 may resolve the ATM address of Router. Router relay the received IP packet toward host B.1. When the IP packet flow from host A.1 to host B.1 is either active flow or connection oriented IP transport flow, the IP packet will be forwarded without IP forwarding process in the router. +----------+ A | Host_A.1 | | +----------+ | | | +---------------+ Subnet | | Switch node | A.0 | +---------------+ | | | +-------------------------+ V | +-------+ +---------+ | | | cell | | IP | | <------------- Router A | | relay +---+ forward | | | | +-------+ +---------+ | | +-------------------------+ | | | +---------------+ Subbet | | Switch node | B.0 | +---------------+ | | | +---------------+ +----------+ | | Switch node |-----| Host_B.1 | | +---------------+ +----------+ | | | +-------------------------+ V | +-------+ +---------+ | | | cell | | IP | | <------------- Router | | relay +---+ forward | | | +-------+ +---------+ | +-------------------------+ | (FIG.3 COIP model) ^^^^^^^^^^^^^^^^^ 3.4. Router Interconnection model Data-link network segment is identical to a single switching node and also is identical to IP subnet. Therefore, IP packet beyond IP subnet must always pass router (i.e. switching node). In order to transfer the IP packet (both connection oriented IP packet and connectionless IP packet) from host A.1 to host Esaki, Nagami & Ohta Expires on April. 20, 1995 [Page 8] INTERNET DRAFT CO&CL IP Forwarding Over ATM Network October1994 C.1, host A.1 transfer IP packet to the router (i.e. switching node) whose subnet ID is A.0. Routers (A.0, B.0 and C.0) relay the received IP packet toward host C.1. When the IP packet flow from host A.1 to host C.1 is either active flow or connection oriented IP transport flow, the IP packet will be forwarded without IP forwarding process in the router. < Subnet A.0 > < Subnet B.0 > < Subnet C.0 > +----------------+ +----------------+ +----------------+ | +------------+ | | +------------+ | | +------------+ | | | IP forward | | | | IP forward | | | | IP forward | | | +-----+------+ | | +-----+------+ | | +-----+------+ | | | | | | | | | | | +-----+------+ | | +-----+------+ | | +-----+------+ | -----+ | cell relay | +-----+ | cell relay | +-----+ | cell relay | +--- | +------------+ | | +------------+ | | +------------+ | +----------------+ +----------------+ +----------------+ / \ / \ +----------+ +----------+ | Host_A.1 | | Host_C.1 | +----------+ +----------+ (FIG.4 Router Interconnection model) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 4. Connection Oriented Service Architecture ------------------------------------------- Currently, the connection oriented IP transport flow (i.e. network layer connection oriented service or sometime said as resource reservation oriented service) has been discussed in IETF, e.g. [RSVP][SIPP1]. The goal of connection oriented communications is to provide end-to-end IP transport. Such a transport flow can have a certain QOS. In such IP transport flow, there is an admission policy at network layer level, that is similar concept to CAC (Connection Admission Control) in ATM network. When the IP transport flow set up request is admitted, the resource for the transport flow is reserved at corresponding routers. Here, in order to find out the next network (or can be said router) that the IP transport flow should be routed, some kind of routing protocol (e.g. similar to OSPF) is executed among the routers. The architecture proposed in this section assumes QOS-ed communication over not only ATM but also other types of platform, using the concept of connection oriented IP transport. In each router the connection oriented IP transport is mapped into data-link layer connection (ATM-VCC) provided by the ATM networks. Connection oriented data-link network (e.g.ATM) can provide QOS-ed connection between any data-link SAPs (Service Access Point) Esaki, Nagami & Ohta Expires on April. 20, 1995 [Page 9] INTERNET DRAFT CO&CL IP Forwarding Over ATM Network October1994 within the data-link network segment through connection set up procedure. Communication beyond data-link network segment is performed by QOS-ed IP transport flow to the router. When the data-link network segment is connection oriented, router or source end-station establishes QOS-ed data-link connection (e.g. ATM-VCC) toward the router. Router performs the same procedure, until the IP transport flow setting up message is reached at the destination end-station. After the IP transport flow is established, IP packets are forwarded through the QOS-ed pipe. A unique identifier, called flow-ID, could be used along the IP transport flow [SIPP1][RSVP]. Within the ATM network, VCI/VPI can be used instead of flow-ID defined in IP layer. The IP packet forwarding procedure is as follows. (1) Packet arrives at router. (2) Flow-ID and IP address (source/destination) are checked and next hop router is determined. (3) TTL (Time To Live) is decremented. (4) Unless the router is destination end-station, packet is forwarded. This procedure is not ATM specific and it is applicable to all other platform. Three cases are discussed below. I. ATM-LAN --> ATM-LAN Router has mapping table between ingress VCI/VPI and egress VCI/VPI for each cell flow according to the appropriate IP transport flow. Then, it is unnecessary to examine the flow-ID and IP address in step (2) : "Re-direction" (concatenation) of ingress and egress data-link connection. Here, mapping table is established during connection set up procedure. (1) Cell arrives at router (2) VCI/VPI of ingress cell is examined and the next hop VCI/VPI of egress cell is determined (maybe by hardware) (3) TTL in IP header may not be decremented (4) Unless the router is destination end-station, cell is forwarded. II. ATM-LAN --> Other-LAN Router will just reassemble IP packet to forward it to the appropriate router or end-station. (1) Cell arrives at router (2) IP packet is reassembled (e.g. AAL5) (3) Flow-ID and IP address (source/destination) is examined and the next hop router is determined. (4) TTL in IP header is decremented (5) Unless the router is destination end-station, packet is forwarded. Esaki, Nagami & Ohta Expires on April. 20, 1995 [Page 10] INTERNET DRAFT CO&CL IP Forwarding Over ATM Network October1994 III. Other-LAN --> ATM-LAN Router has mapping table between ingress flow-ID/IP-address and egress VCI/VPI for each IP transport flow. Here, again, the mapping table is established during connection set up procedure. Here, the TTL decrement value in step 3 may not be one. This because, within the ATM-LANs, the TTL of IP packet header is not dealt with at the bypassed router, even when the IP packet passes through the router. (1) Packet arrives at router (2) Flow-ID and IP address (source/destination) is examined and the next hop VCI/VPI of egress cell is determined. (3) TTL in IP header is decremented (may not be one) When ATM end-stations communicates purely over ATM-LANs, a seamless end-to-end cell-relaying virtual channel can be established between them. It should be noted that, even though many ATM-VCCs (data-link connection) are concatenated at data-link layer, the seamless connection over ATM-LANs is at network layer connection (not data-link layer connection). 5. Connectionless Service Architecture -------------------------------------- 5.1. Connectionless Service over ATM Networks Connection oriented service must perform IP level connection set-up procedure, before transferring the information. It is said that connection set up latency for a large scaled ATM network would be small [Schmt]. On the contrary, connectionless service does not require IP level connection set up procedure, basically. ATM-VCC (data-link connection), in order to provide connectionless service, has been established in advance. Here, ATM-VCCs to provide connectionless (conventional IP forwarding) service would be SPVC (Semi-Permanent VC) or SVC (Switched VC), rather than PVC (Permanent VC). Once the VC(s) for connectionless service is(are) established, there will be no connection set up latency to transfer connectionless IP packet. Therefore, even when ATM networks and other networks provide end-to-end connection oriented data transmission service, connectionless service will be still important. And, from the view point of co-existence with the data-link network that is not connection oriented (e.g. FFOL), connectionless data transmission service (i.e. conventional IP forwarding) must be provided by the ATM networks. 5.2. Connectionless Services in Conventional Data Networks In IP networks, IP processing will be generally performed, whenever the IP packet is transferred to other data-link network segment [Comer]. This means that, in a large scaled network, the IP packet will experience many IP processing points between source and destination end-stations. Therefore, in the large scaled network, the large delay for IP packet delivery is expected, due to large number of IP processing points. This is similar to CLS Esaki, Nagami & Ohta Expires on April. 20, 1995 [Page 11] INTERNET DRAFT CO&CL IP Forwarding Over ATM Network October1994 (ConnectionLess Server) approach defined by ITU-T [I.364]. Each router can directly recognize the destination subnet (i.e.Net-ID) of the received IP packet, as the address view of routing domain. However, the network layer protocol (e.g. IP) is performed at every router to terminate data-link layer (i.e. MAC). As a result, even the distance associated with the address management view is small, the distance associated with the data-link layer's view is not small. 5.3. Proposed Connectionless Service Architecture Connectionless IP packet transmission in the proposed architecture is fundamentally no different from that of existing IP forwarding. Within the data-link network segment, end-station and routers exchange the IP packet using data-link connection (e.g.ATM-VCC). IP packet beyond the data-link segment is sent to a appropriate router from source end-station. If a router has some knowledge on local topology (e.g. topology information in routing domain) that it can determine not only the next but also the second or third next hops, it is possible to bypass IP layer processing on the intermediate routers, by having direct ATM connection between distant routers. Routers exchange routing information and forward IP packets to the appropriate router. Following discussion is regarding the network that has only ATM-LANs. Regarding toward the other types of platform from ATM network, ATM-VCC shall be always terminated at the router to execute conventional IP forwarding process. On the contrary, regarding toward ATM network from the other types of platform, you can see the router as the end-station in the following discussion. The proposed ATM router has both cell relaying and IP forwarding capabilities. IP packet toward the out-side ATM-LAN where the source end-station belongs to is sent to a appropriate router from the source end-station. Routers exchange routing information and forward packets to the appropriate router. Regarding the current routing protocol (e.g. OSPF), every router does know all of subnet IDs (Net-IDs) within the domain where the routing protocol is applied. 5.3.1. Hop-by-hop ATM-VCC cacheing for active IP flow When the first (of session or of after idle period) IP packet arrives at router (attaching to ATM), the next router is decided using IP header, i.e. route decision will be done based on IP address (and flow-ID in IPv6). After certain next router is decided, a dedicated ATM-VCC is picked up for the IP packet flow. When the packet is come from ATM (or connection-oriented platform, e.g. Frame Relay), router maps between incoming VCI/VPI (or some flow-ID in data-link layer)and (picked up) outgoing VCI/VPI. When the packet is coming from the conventional platform (e.g. Ethernet/FDDI), router always examine IP header in the case of IPv4 (in IPv6, flow-ID could be mapped with outgoing VCI/VPI). When IP packet is come from ATM, the successive IP packet is forwarded based on incoming VCI/VPI value, without any usual IP processing. Esaki, Nagami & Ohta Expires on April. 20, 1995 [Page 12] INTERNET DRAFT CO&CL IP Forwarding Over ATM Network October1994 If the activity of IP packet flow is/becomes low, the mapping information (i.e. cached information) is pushed out. That means that the pushed out IP packet flow from cache must be examined usual IP processing (looking at IP header) at the router. As a result, high activity IP packet flow can bypass a usual IP processing at the router. Above procedure is performed at every router (e.g. intermediate router that interconnects ATM networks). Then, intermediate router can deal with high activity IP packet flow with very small latency. This is because IP packet forwarding is done by cell relaying, which will be done by hardware. The job of router is maintaining VCI/VPI mapping table in the cell switch, or is maintaining mapping table between flow-ID and VCI/VPI, as a cache (bypass) processing path for usual IP forwarding process. When every data-link segment is ATM, small latent IP packet forwarding will be provided, even for connectionless IP packet. This is because IP packet is relayed by cell-by-cell at every routers, regarding active IP packet flow. 5.3.2. Bypassed ATM VCC for active transaction When routers knows that it will have or is having a lot of communication with a certain router (or subnet), it may create bypass ATM connection (ATM-VCC) which is seamlessly connects the two routers. This bypasses ATM-VCC cuts through the IP processing at the intermediate routers along the cell transmission route between the two routers. On the other hand, when router does not use bypassed ATM-VCC for datagram forwarding, the datagram will be forwarded hop-by-hop way, that is completely same approach as the current IP forwarding. When router can establish a seamless ATM-VCC bypassing the intermediate router's network layer protocol processing along the routing path to the destination end-station, we can reduce the end-to-end datagram delivery latency due to the reducing the number of network layer protocol processing points at the intermediate routers. Here, it is unnecessary to propagate the information of such bypassed ATM-VCC for every routers. The information of bypassed ATM-VCC shall be maintained by the source and destination routers associated with the bypassed ATM-VCCs. This means that the routing protocol and routing information exchanged among routers is nothing different from the existing routing protocol. The information and path regarding bypassed IP forwarding process at the intermediate router(s) could be seen as a cache image associated with the information and path for hop-by-hop route (i.e. conventional IP forwarding route). When the cache is missed (i.e. when there is no bypass route), the hop-by-hop information is referenced and the IP packet is forwarded to the adjacent router. On the contrary, when the cache is hit (i.e. when a bypass route exists associated with the destination subnet), the cache information is referenced and the IP packet is forwarded using the bypass route. The establishment of bypass route will be issued by the Esaki, Nagami & Ohta Expires on April. 20, 1995 [Page 13] INTERNET DRAFT CO&CL IP Forwarding Over ATM Network October1994 following two cases. Here, source router and destination router used in this subsection can be intermediate router, from the view point of end-stations. This means that source (or destination) router may not directly connected the source (or destination) end-station through a single ATM-VCC. In other words, in order to reach source router from source end-station (or to reach destination end-station from destination router), intermediate router(s) may be required. (1) Source router Source router has activity information regarding the destination subnet's. When the IP forwarding activity toward the certain destination subnet is high, the source router establishes the bypass ATM-VCC toward the destination subnet's router. When the activity becomes low, the established bypass ATM-VCC will be teared down (i.e. cache information is pushed out). Bypassed ATM-VCC from the source router toward the destination router can be established without activity information. This means that bypass ATM-VCC can be established both dynamically and statistically based on network's configurational information. * Here, NBMA net's short-cut routing can be realized when we assume that the creating bypass route is performed at ATM-VCC establishment phase (i.e. signaling phase) for connection oriented IP service or when we assume that the creating bypass route is performed during address resolution procedure for connectionless IP service. These procedure can be analogous to "re-direction" procedure in RFC1620, or can be analogous to ARP procedure in NHRP [RFC1620][NHRP]. (2) Intermediate router (cell-relay re-direction) When the intermediate router becomes high load, the intermediate router can create bypass routes that is bypassing itself. The intermediate router establishes the bypass ATM-VCC for active source and destination router pair. After the establishment of two ATM-VCCs, that are (a) from source router to intermediate router and (b) from intermediate router to destination router, these two ATM-VCCs are coupled at the intermediate router. This procedure could be seen as re-direction of cell-relaying : i.e. the cell-relaying coming from the source router to IP forwarding entity in the intermediate router is "re-directed" to the cell-relaying entity toward the destination router. The ideal and optimal case is full meshed ATM-VCCs are established among all of routers that are exists in the routing entry of routing information. Obviously, the every host can act as same as the router discussed above (i.e. you can see the router as a end-station). Then, the Esaki, Nagami & Ohta Expires on April. 20, 1995 [Page 14] INTERNET DRAFT CO&CL IP Forwarding Over ATM Network October1994 seamless cell transmission channel will be established, i.e. (a) from source end-station to router, (b) from source end-station to destination end-station, and (c) from router to destination end-station. 6. Summary ---------- High throughput small latent datagram (IP packet) delivery architecture using ATM technology is proposed in this contribution. Using the ATM's cell relaying capability at router, we can provide high throughput small latent end-to-end IP packet delivery. The proposed architecture can be generally applied to all of network reference model discussed in IETF and ATM Forum. Regarding connection oriented IP packet transmission, since the router can map between flow-ID (and IP address) in the IP packet header and VCI/VPI, we can reduce the IP processing latency at the routes within the ATM network. Regarding connectionless IP packet transmission, we can also reduce the IP processing latency at the routers within the ATM networks, through the re-direction (bypassing) technique at the routers. When the proposed architecture is applied, we can reduce the processing load of each router, as well as reducing end-to-end packet delivery delay. This is because, due to the bypassing IP processing by re-direct technique, the amount of IP processing for intermediate IP packets can be reduced. Acknowledgement : The authors thanks to many ATM experts in JAIN consortium in Japan for many discussion regarding the proposed architecture in this paper. And, the authors appreciate many discussions with Mr.Oumar Ndiaye, who belonged to MIT in USA, regarding a primary consideration on the software implementation of the proposed architecture. Esaki, Nagami & Ohta Expires on April. 20, 1995 [Page 15] INTERNET DRAFT CO&CL IP Forwarding Over ATM Network October1994 References ---------- [I.150] ITU-T Rec.I.150 : "BISDN Asynchronous Transfer Mode", 1990 [I.364] ITU-T Rec.I.364 : "Support of Broadband Connectionless Data Service on BISDN", June, 1992 [ATM Forum] ATM Forum : "ATM UNI Specification 3.0", Aug.1993 [Schmit] A.Schmidt, R.Campbell : "Internet Protocol Traffic Analysis with Applications for ATM Switch Design", Computer Communications Review, vol.23, No.2, pp.39-52, April, 1993 [CLIP] M.Laubach ;"Classical IP and ARP over ATM", RFC1577, Jan., 1994. [NBMA] J.Heinanen ;"NBMA Address Resolution Protocol", (Internet-Draft), draft-ietf-atm-nbma-01.txt, June, 1993 [NHRP] D.Katz, D.Piscitello ;"NBMA Next Hop Resolution Protocol (NHRP)", (Internet-Draft), draft-ietf-rolc-nhrp-02.txt, Aug., 1994 [RFC1620] B.Braden, J.Postel, Y.Rekhter, "Internet Architecture Extensions for Shared Media", RFC 1620, May, 1994. [SIPP1] S.Deering : "Simple Internet Protocol Plus (SIPP) Specification (128-bit address version)", (Internet-Draft), draft-ietf-sipp-spec-01.txt, July, 1994 [SIPP2] P.Francis, S.Deering, R.Hinden, R.Govindan : "Simple Internet Protocol Plus (SIPP) : Addressing Architecture", (Internet-Draft), draft-ietf-sipp-routing-addr-02.txt, July, 1994 [IPng] F.Kastenholz, C.Partridge : "Technical Criteria for Choosing IP:The Next Generation (IPng)" (Internet-Draft), draft-kastenholz-ipng-criteria-02.txt, May, 1994 [RSVP] L.Zhang ;"Resource ReSerVation Protocol (RSVP)", (Internet-Draft), draft-ietf-rsvp-spec-02.ps, May, 1994 [Ohta] M.Ohta, H.Esaki, K.Nagami ;"Conventional IP over ATM", (Internet-Draft) draft-ohta-ip-over-atm-01.txt, July, 1994 [Comer] D.E.Comer : "Internetworking with TCP/IP", Prentice Hall, 1991 [Esaki1] H.Esaki, Y.Tsuda, T.Saito, S.Natsubori : "Class D Service Architecture in ATM-Internet", ICC'94, May, 1994. [Esaki2] H.Esaki, Y.Tsuda, T.Saito, S.Natsubori : "Datagram Delivery in an ATM-Internet", IEICE Trans. on Communications, Special Issues for Future Private Networks, March, 1994 Esaki, Nagami & Ohta Expires on April. 20, 1995 [Page 16] INTERNET DRAFT CO&CL IP Forwarding Over ATM Network October1994 Authors' Addresses Hiroshi Esaki R&D Center, Toshiba Corporation 1 Komukai-Toshiba-cho, Saiwai-ku Kawasaki 210, JAPAN Phone: +81-44-549-2238 Fax: +81-44-549-2262 EMail: hiroshi@csl.rdc.toshiba.co.jp Ken-ichi Nagami R&D Center, Toshiba Corporation 1 Komukai-Toshiba-cho, Saiwai-ku Kawasaki 210, JAPAN Phone: +81-44-549-2238 Fax: +81-44-549-2262 EMail: nagami@csl.rdc.toshiba.co.jp Masataka Ohta Computer Center Tokyo Institute of Technology 2-12-1, O-okayama, Meguro-ku Tokyo 152, JAPAN Phone: +81-3-5434-3299 Fax: +81-3-5734-3415 EMail: mohta@necom830.cc.titech.ac.jp Esaki, Nagami & Ohta Expires on April. 20, 1995 [Page 17]