Network Working Group Jerry Ash Internet Draft Bur Goode Jim Hand Expiration Date: March 2004 AT&T Raymond Zhang Infonet Services Corporation October, 2003 Requirements for VoIP Header Compression over Multiple-Hop Paths Status of this Memo This document is an Internet-Draft and is in full conformance with all provisions of Section 10 of RFC2026. 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. ABSTRACT: VoIP typically uses the encapsulation voice/RTP/UDP/IP/. When MPLS labels are added, this becomes voice/RTP/UDP/IP/MPLS-labels. For an MPLS VPN, the packet header is at least 48 bytes, while the voice payload is often no more than 30 bytes. VoIP header compression can significantly reduce the VoIP overhead through various compression mechanisms. This is important on access links where bandwidth is scarce, and can be important on backbone facilities, especially where costs are high (e.g., some global cross-sections). Providing VoIP header compression over multiple hop paths avoids multiple compression/decompression cycles at each router, and thereby increases the processing scalability of the maximum number of simultaneous VoIP flows which use header compression at each router. This draft gives a problem statement, example scenarios, goals, and requirements for VoIP header compression from compressor to decompressor over multiple-hop paths, possibly over MPLS. It also discusses work items and issues for consideration, and highlights various solution alternatives that have been proposed. Table of Contents 1. Introduction 2. Example Scenarios 3. Problem Statement, Goals, & Requirements 4. Work Items & Issues 5. Security Considerations 6. References 7. Authors' Addresses Appendix A. Summary of Proposed Solutions 8. Full Copyright Statement 1. Introduction Voice over IP (VoIP) typically uses the encapsulation voice/RTP/UDP/IP/. When MPLS labels are added, this becomes voice/RTP/UDP/IP/MPLS-labels. For an MPLS VPN (e.g., [MPLS-VPN], the packet header is at least 48 bytes, while the voice payload is often no more than 30 bytes. The interest in VoIP header compression is to exploit the possibility of significantly reducing the VoIP overhead through various compression mechanisms, and also to increase scalability of VoIP header compression. Providing VoIP header compression over multiple hop paths avoids multiple compression/decompression cycles at each router, and thereby increases the processing scalability of the maximum number of simultaneous VoIP flows which use header compression at each router. The need for compression may be important on access links where bandwidth is more scarce, but it could be important on backbone facilities, especially where costs are high (e.g., some global cross-sections). Typically on access and international routes, VoIP will use voice compression mechanisms (e.g., G.729) in order to conserve bandwidth. With VoIP header compression, significantly more bandwidth could be saved. For example, carrying VoIP headers for the entire voice load of a large domestic network with 300 million or more calls per day could consume on the order of about 20-40 gigabits-per-second on the backbone network for headers alone. This overhead could translate into considerable bandwidth capacity. The claim is often made that once fiber is in place, increasing the bandwidth capacity is inexpensive, nearly 'free'. This may be true in some cases, however, on some international cross-sections, especially, facility/transport costs are very high and saving bandwidth on such backbone links is very worthwhile. Decreasing the backbone bandwidth is needed in some areas of the world where bandwidth is very expensive. More important in almost all locations is decreasing the access bandwidth. While VoIP header compression over multiple hops clearly helps decrease bandwidth requirements, it also avoids compression-decompression cycles at access boundaries. So although bandwidth is getting cheaper, the value of compression does not go away. In addition, IPv6 will increase the size of headers and therefore increase the importance of header compression for VoIP flows. In principle, we could use existing header compression techniques, such as those described in [cRTP], together with MPLS labels to make the transport of the RTP/UDP/IP headers more efficient over an MPLS network. Alternatively, if MPLS labels are not used for forwarding, mechanisms based on a session context ID (SCID) could be used to route compressed packets from compressor to decompressor over multiple hop paths. However, at this time, there are no standards for VoIP header compression from compressor to decompressor over multiple hop paths, and vendors have not implemented such techniques. [cRTP] header compression is often used on the access links from the CE router to the PE router. However, cRTP header compression must be implemented on a hop-by-hop basis, and does not scale well for large voice traffic loads. The maximum number of cRTP flows is about 30 for a typical customer premise router, but the need can exceed 300 for a high-end case, and 300-500 for a typical PE router. By using VoIP header compression over multiple hops, routers merely forward compressed packets without doing a decompression/recompression cycle, thereby increasing the maximum number of simultaneous VoIP compressed flows that routers can handle. To implement header compression over multiple hops, the ingress router/gateway would have to apply the header compression to the IP packet, the compressed packet routed over multiple hops using MPLS labels, SCID switching, or some other method, and the compressed header would have to be decompressed at the egress router/gateway where the multiple hop VoIP header compression session terminates. A method is needed to set up a correspondence between the header compression tables at the ingress and egress of the multiple hop VoIP header compression session. VoIP over MPLS (VoMPLS) header compression methods have been proposed earlier in the MPLS working group, however the relevant drafts have expired. George Swallow and Lou Berger proposed a 'simple' end-to-end compression scheme in which all first-order differences in the RTP/UDP/IP headers were transmitted, and in which the header compression context was established through RSVP signaling [RSVP, RSVP-TE]. Swallow's and Berger's 'simple' approach achieved about a 50 percent reduction in the size of the RTP/UDP/IP header. Thomas Theimer proposed an end-to-end header compression approach that would provide MPLS extensions for tunneling compressed headers over PPP links [TCRTP]. Lou Berger proposed MPLS-based extensions to 'hop-by-hop' header compression methods (e.g., [cRTP]), which could achieve about 90 percent reduction in the RTP/UDP/IP header. The MPLS Forum has since finalized an implementation/protocol for VoMPLS header compression [MPLSF-HC], however the technique does not provide means to interwork with IP. This draft gives a problem statement, example scenarios, goals, and requirements for VoIP header compression from compressor to decompressor over multiple-hop paths, possibly over MPLS. It also discusses work items and issues for consideration, and highlights various solution alternatives that have been proposed. In Section 2 we give example scenarios, in Section 3 we give a problem statement, goals, and requirements, in Section 4 we discuss work items and issues for consideration, and in Appendix A we highlight various solution alternatives that have been proposed. 2. Example Scenarios Scenario A: Small customer sites with IP phones or VoIP terminal adapters connect to a CE routers serving as header compression gateways. A VoIP session is established from CE1/HC --> PE1 --> P --> PE2 --> CE2/HD, where CE1/HC is the customer edge (CE) router where header compression (HC) is performed, and CE2/HD is the CE router where header decompression (HD) is done. The voice traffic from CE1/HC to CE2/HD is typically small, on the order of only a few simultaneous calls in peak periods. For example, cRTP compression of the RTP/UDP/IP header is performed at CE1/HD, and the compressed packets are routed from CE1/HC to PE1, P, PE2, and finally to CE2/HD, without further decompression/recompression cycles. The compressed packets might be routed using MPLS labels or perhaps use SCID switching to route the compressed packets from compressor CE1/HC to decompressor CE2/HD over the multiple hop path. The RTP/UDP/IP header is decompressed at CE2/HD and can be forwarded to other routers, as needed. In the example scenario, cRTP header compression therefore takes place between the end-points, and in the case of an MPLS path, the MPLS path appears as a single link-layer to the compressor, even though it may traverse several routers. The MPLS path transports cRTP/MPLS-labels instead of RTP/UDP/IP/MPLS-labels, saving 36 octets per packet. The MPLS label stack and link-layer headers are not compressed. Additional signaling is needed to map the context for the compressed packets. It is desirable that the packet loss rate between CE1/HC and CE2/HD not exceed 0.001, that packet delay variation not exceed 50 ms., and that packet transfer delay not exceed 400 ms. Scenario B: Many VoIP flows are originated from customer sites such as CE1/HC to several large customer call centers served by PE2, which include CE2/HD, CE3/HD, and CE4/HD. It is essential that the PE2-CE2/HD, PE2-CE3/HD, and PE2-CE3/HD hops all use header compression to allow a maximum number of simultaneous VoIP flows. To allow processing at PE2 to handle the volume of simultaneous VoIP flows, it is desired to use multi-hop header compression for these flows. That is, with multi-hop header compression, PE2 does not need to do header compression/decompression for the flows to the call centers, enabling more scalability of the number of simultaneous VoIP flows with header compression at PE2. 3. Problem Statement, Goals, & Requirements VoIP typically uses the encapsulation voice/RTP/UDP/IP/. When MPLS labels are added, this becomes voice/RTP/UDP/IP/MPLS. Typically, VoIP will use voice compression mechanisms (e.g., G.729) in order to conserve bandwidth. For an MPLS VPN, the packet header is at least 48 bytes, while the compressed voice payload is typically no more than 30 bytes. With VoIP header compression, significantly more bandwidth could be saved. Therefore, VoIP header compression from compressor to decompressor over multiple-hop paths, possibly over MPLS, is required in order to significantly reduce the VoIP overhead through compression mechanisms. The goals of VoIP header compression over multiple-hop paths are as follows: a. provide more efficient voice transport, b. increase the scalability of VoIP header compression to a very large number of flows, c. not significantly degrade packet delay, delay variation, or loss probability, and d. allow efficient signaling of header context from compressor to decompressor. The requirements for VoIP header compression from compressor to decompressor over multiple-hop paths, possibly over MPLS, are that it MUST: a. allow VoIP header compression from compressor to decompressor over multiple-hop paths, possibly through an MPLS network, and thereby avoid hop-by-hop compression/decompression cycles ([E2E-VoMPLS], [E2E-cRTP]). b. compress RTP/UDP/IP headers by at least 50%, to provide for efficient voice transport. c. allow VoIP header compression over multiple-hop paths with delay not to exceed 400 ms. from compressor to decompressor [Y.1541, G.114]. d. allow VoIP header compression over multiple-hop paths with delay variation not to exceed 50 ms. from compressor to decompressor [Y.1541, G.114]. e. allow VoIP header compression over multiple-hop paths with packet loss not to exceed 0.001 from compressor to decompressor [Y.1541, G.114]. f. support various voice encoding supported by [RTP] (G.729, G.723.1, etc.). g. be scalable to up to 20,000 simultaneous flows (e.g., PE/HC --> PE/HD). h. use standard compress/decompress algorithms (e.g., [cRTP], [ECRTP], [ROHC]) to compress the RTP/UDP/IP headers. i. allow use of standard protocols to make [cRTP] more tolerant of packet loss (e.g., [ECRTP]). j. operate in non-MPLS networks (i.e., without use of MPLS labels). k. operate in MPLS [MPLS-ARCH] networks, with use of MPLS labels, using either [LDP] or [RSVP] signaling. l. operate in RFC2547 VPN context [MPLS-VPN]. m. allow use of standard protocols to signal context identification and control information (e.g., [RSVP], [RSVP-TE], [LDP]). n. use standard protocols to aggregate RSVP-TE signaling (e.g., [RSVP-AGG]}. o. allow use of standard protocols to prioritize packets (e.g., [DIFFSERV, DIFF-MPLS]). p. allow use of standard protocols to allocate LSP bandwidth (e.g., [DS-TE]). The solution space is discussed in Appendix A, and includes the well-established cRTP-based [cRTP] and enhanced cRTP [ECRTP] approaches, as specified in RFC 2508 and RFC 3545, respectively, 'Simple' header compression [SIMPLE] and the approaches described in [E2E-VoMPLS] and [E2E-cRTP]. 4. Work Items & Issues Based on the above requirements, there are several work items to be considered: 1. Extend cRTP to work from compressor to decompressor over multiple-hop paths with moderate delay (e.g., < 400 ms.) and moderate packet loss (e.g., < 2%). We will assume enhanced cRTP [ECRTP] is sufficient for now. 2. How to directly route cRTP packets using SCID switching rather than doing a decompression/routing/compression cycle. SCID switching is independent of MPLS, and a router can do this in isolation, without being observable by upstream or downstream routers. It will look to cRTP as a 'link' with higher latency. 3. How to do ECRTP over an MPLS LSP, in which new RSVP signaling extensions are needed. Here the compression between ingress-egress routers of an LSP can be viewed as the MPLS equivalent of RFC 2509 4. How SCID switching can be applied by the ingress router of a compressed MPLS LSP. This in effect combines the outputs of work items 2 and 3. There are several issues to be considered within these work items: 1. protocol extensions for cRTP, RSVP-TE, RFC2547 VPNs 2. resynchronization & performance of cRTP/'simple' mechanisms 3. scalability of E2E VoMPLS applied CE-CE 4. LDP application as the underlying LSP signaling mechanism Protocol extensions may be required for [cRTP] and [ECRTP] in that a packet type field is needed to identify FULL_HEADER, CONTEXT_STATE, etc. packets, and mechanisms are needed to create 'SCID routing tables' to allow routing based on the session context ID (SCID). New objects need to be defined for [RSVP-TE], and extensions to RFC2547 VPNs are needed to allow SCID routing combined with label switching at PE's. These protocol extensions need coordination with other working groups (MPLS, L3/2VPN, etc.). Resynchronization and performance of VoIP header compression over multiple-hop paths needs to be considered. cRTP header compression might not perform well with frequent resynchronizations, so standard protocols like [ECRTP] should be used to make cRTP more tolerant of packet loss. The [SIMPLE] approach avoids the need for resynchronization, however cRTP achieves greater efficiency than 'simple' (90% vs. 50% header compression), but requires resynchronization. [ECRTP] minimizes feedback based error recovery using CONTEXT_STATE packets to make cRTP more tolerant of packet loss, and minimize the need to use the cRTP error recovery mechanism. cRTP performs best with very low packet error rates on all hops of the path. Tunneling a cRTP session through multiple IP hops will increase the round trip delay and the chance of packet loss. cRTP contexts are invalidated due to packet loss. The cRTP error recovery mechanism using CONTEXT_STATE packets can compound the problem when long round trip delays are involved. When the cRTP decompressor context state gets out of synch with the compressor, it will drop packets associated with the context until the two states are resynchronized. To resynchronize context state at the two ends, the decompressor transmits the CONTEXT_STATE packet to the compressor, and the compressor transmits a FULL_HEADER packet to the decompressor. If the resynchronization were rare, then the basic cRTP approach would perform well for end-to-end header compression. Otherwise, ECRTP is desirable. Scalability of VoIP header compression from compressor to decompressor over MPLS multiple-hop paths, needs to be considered when applied CE/HC to CE/HD. This requires that LSPs be established with RSVP-TE between many CE pairs, raising possible scalability issues. RSVP-TE has advantages in that it allows VoIP bandwidth assignment on LSPs and has QoS mechanisms. But if applied CE/HC-CE/HD it may require a large number of LSPs to be created, and there is concern for CE, PE, and P ability to do the necessary processing. This includes: a. processing and label binding at every MPLS node on every path between each CE/HC-CE/HD pair, b. processing every time resource reservation is modified (e.g., to adjust to varying number of calls on a CE/HC-CE/HD pair), and c. core router load from thousands of LSPs, setup commands, refresh messages. A final issue concerns LDP application as the underlying LSP signaling mechanism. It is desirable to signal VoIP over MPLS tunnels with LDP, since many RFC2547 VPN implementations use LDP as underlying LSP signaling mechanism, and LDP is very scalable. However, substantial extensions to LDP may be needed to signal VoIP over MPLS header compression over multiple-hop paths. [LDP-PWE3] proposes ways for LDP to signal 'VC' (outer) labels for pseudo wires, and also suggests ways to do auto-discovery. This together with LDP capability to distribute inner labels might support CE/HC-CE/HD VoIP header compression LSPs within the context of RFC 2547. Also, [GVPLS] proposes LDP signaling for multipoint-to-point LSPs. However, LDP provides no bandwidth associated with LSPs, and QoS mechanisms are limited. 5. Security Considerations No new requirements. 6. References [cRTP] Casner, S., Jacobsen, V., "Compressing IP/UDP/RTP Headers for Low-Speed Serial Links", RFC 2508, February 1999. [DIFF-MPLS] Le Faucheur, F., et. al., "MPLS Support of Diff-Serv", RFC 3270, May 2002. [DIFFSERV] Blake, S., et. al., "An Architecture for Differentiated Services", RFC 2475, December 1998. [DS-TE] Le Faucheur, F., Lai, W., et. al., "Requirements for support of Diff-Serv-aware MPLS Traffic Engineering," RFC 3564, July 2003. [E2E-VoMPLS] Ash, G., Goode, B., Hand, J., "End-to-End VoIP over MPLS Header Compression", work in progress. [E2E-cRTP] Ash, G., Goode, B., Hand, J., "End-to-End VoIP Header Compression Using cRTP", work in progress. [ECRTP] Koren, T., et. al., "Compressing IP/UDP/RTP Headers on Links with High Delay, Packet Loss, and Reordering," RFC 3545, July 2003. [G.114] "One-way Transmission Time," January 2003. [GVPLS] Radoaca, V., et. al., "GVPLS/LPE - Generalized VPLS Solution based on LPE Framework," work in progress. [KEY] Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", RFC 2119, March 1997. [LDP] Andersson, L., et. al., "LDP Specification", RFC 3036, January 2001. [LDP-PWE3] Rosen, E., "LDP-based Signaling for L2VPNs", work in progress. [MPLS-ARCH] Rosen, E., et. al., "Multiprotocol Label Switching Architecture," RFC 3031, January 2001. [MPLS-VPN] Rosen, E., Rekhter, Y., "BGP/MPLS VPNs", RFC 2547, March 1999. [MPLSF-HC] MPLS Forum Technical Committee, "Voice over MPLS - BearerTransport Implementation Agreement," March 2001. [ROHC] Bormann, C., et. al., "Robust Header Compression (ROHC)," RFC 3091, July 2001. [RSVP] Braden, R. et al., "Resource ReSerVation Protocol (RSVP) -- Version 1, Functional Specification", RFC 2205, September 1997. [RSVP-AGG] Baker, F., et. al., "Aggregation of RSVP for IPv4 and IPv6 Reservations", RFC 3175, September 2001. [RSVP-TE] Awduche, D., et. al., "RSVP-TE: Extensions to RSVP for LSP Tunnels", RFC 3209, December 2001. [RTP] Casner, S., et. al., "RTP: A Transport Protocol for Real-Time Applications," RFC 3550, July 2003. [SIMPLE] Swallow, G., Berger, L., "Simple Header Compression", work in progress. [TCRTP] Thompson, B., et. al., "Tunneling Multiplexed Compressed RTP ("TCRTP")", work in progress. [Y.1541] "Network performance objectives for IP-based services," December 2002. 7. Authors' Addresses Jerry Ash AT&T Room MT D5-2A01 200 Laurel Avenue Middletown, NJ 07748, USA Phone: +1 732-420-4578 Email: gash@att.com Bur Goode AT&T Phone: + 1 203-341-8705 E-mail: bgoode@att.com Jim Hand AT&T Room MT A2-4F36 200 Laurel Avenue Middletown, NJ 07748, USA Phone: +1 732-420-6179 E-mail: jameshand@att.com Raymond Zhang Infonet Services Corporation 2160 E. Grand Ave. El Segundo, CA 90025 USA Email: raymond_zhang@infonet.com Appendix A - Summary of Proposed Solutions The solution space includes the well-established cRTP-based [cRTP] and enhanced cRTP [ECRTP] approaches, as specified in RFC 2508 and RFC 3545, respectively. These methods presently provide hop-by-hop header compression, and require resynchronization when packets are lost, but are very efficient providing about 90% header compression. 'Simple' header compression [SIMPLE] over multiple-hop MPLS paths transmits only first order differences, and as such resynchronization is not needed when packets are lost. This approach achieves about 50% header compression, and was accepted by the MPLS working group for a charter extension at IETF-47, although the work was discontinued before the charter was extended. The approach described in [E2E-VoMPLS] 'End-to-End VoIP over MPLS Header Compression', has the following steps: 1. use RSVP to establish LSPs between CE1/header compressor (HC) and CE2/header decompressor (HD) 2. use cRTP (or simple) to compress header at CE1/HC, decompress at CE2/HD 3. CE1/HC requests session context IDs (SCIDs) from CE2/HD 4. CE1/HC appends CE2/HD label to compressed header 5. header compression context routed from CE1/HC --> PE1 --> P --> PE2 --> CE2/HD 6. route compressed packets with MPLS labels CE1/HC --> CE2/HD 7. packet decompressed at CE2/HD using cRTP (or simple) algorithm The advantage of this approach is that it minimizes PE requirements, but has the disadvantage that CE's need to run RSVP, which is a possible scalability issue. It also adds a label, which increases the header size by four bytes, thereby reducing the compression efficiency. The approach described in [E2E-cRTP] 'End-to-End VoIP Header Compression Using cRTP' has the following steps: 1. use RSVP to establish LSPs between PE1-PE2 2. use cRTP to compress header at CE1/HC, decompress at CE2/HD 3. PE1 requests SCIDs from PE2 4. header compression context routed from CE1/HC --> PE1 --> P --> PE2 --> CE2/HD 5. PE1 & PE2 create 'SCID routing tables' & perform 'SCID switching' for compressed packets (SCID --> MPLS labels) 6. route compressed packets with MPLS labels PE1 --> PE2 7. packet decompressed at CE2/HD using cRTP algorithm The advantage of this approach is that it minimizes CE requirements and minimizes bandwidth requirements on the access links, but has the disadvantage that there are additional PE requirements, such as the need to create 'SCID routing tables'. 8. Full Copyright Statement Copyright (C) The Internet Society (2003). All Rights Reserved. This document and translations of it may be copied and furnished to others, and derivative works that comment on or otherwise explain it or assist in its implementation may be prepared, copied, published and distributed, in whole or in part, without restriction of any kind, provided that the above copyright notice and this paragraph are included on all such copies and derivative works. However, this document itself may not be modified in any way, such as by removing the copyright notice or references to the Internet Society or other Internet organizations, except as needed for the purpose of developing Internet standards in which case the procedures for copyrights defined in the Internet Standards process must be followed, or as required to translate it into languages other than English. The limited permissions granted above are perpetual and will not be revoked by the Internet Society or its successors or assigns. This document and the information contained herein is provided on an "AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.