Network Working Group Krister Svanbro, Ericsson INTERNET-DRAFT Sweden Expires: April 2001 October 10, 2000 Lower Layer Guidelines for Robust RTP/UDP/IP Header Compression 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 cite them other than as "work in progress". The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/lid-abstracts.txt The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html This document is a submission of the IETF ROHC WG. Comments should be directed to its mailing list, rohc@cdt.luth.se. Abstract This document describes lower layer guidelines for robust header compression (ROHC) and the requirements ROHC put on lower layers. The purpose of this document is to support the incorporation of robust header compression algorithms, as specified in the ROHC working group, into different systems such as those specified by 3GPP, 3GPP2, ETSI, etc. The document covers only lower layer guidelines for compression of RTP/UDP/IP and UDP/IP headers as specified in [ROHC]. Both general guidelines and guidelines specific for cellular systems are treated. Svanbro [Page 1] INTERNET-DRAFT Lower Layer Guidelines for Robust HC October 10, 2000 TABLE OF CONTENTS 1. Introduction..................................................3 2. Terminology...................................................3 3. General guidelines............................................4 3.1 Error detection.......................................4 3.2 Inferred header field information.....................4 3.3 Handling of header size variation.....................5 3.4 Negotiation of header compression parameters..........6 3.5 Demultiplexing of flows onto logically separated channels..................................6 3.6 Packet type identification............................6 3.7 Packet duplication....................................6 3.8 Feedback packets......................................7 4. Cellular system specific guidelines...........................7 4.1 Handover procedures...................................7 4.2 Unequal error detection...............................8 4.3 Unequal error protection..............................9 5. Author's addresses............................................9 6. References...................................................10 Svanbro [Page 2] INTERNET-DRAFT Lower Layer Guidelines for Robust HC October 10, 2000 1. Introduction Almost all header compression algorithms [RFC1144, RFC2507, RFC2508] rely on some functionality from underlying link layer. Headers (compressed or not) are expected to be delivered without any residual bit errors, IP length fields are inferred from link layer length fields and packet type identification may be separated from the header compression scheme and instead done at underlying link layer. [RFC2509], for example, elaborates on how to incorporate IP header compression [RFC2507] in PPP [RFC1661]. It is important to be aware of such assumptions on required functionality from underlying layers when incorporating a header compression scheme into a system. The functionality required by a specific header compression scheme from lower layers may also be needed if incorporation of a header compression scheme is to be prepared without knowing the exact details of the final scheme. This document describes lower layer guidelines for robust RTP/UDP/IP header compression [ROHC] as being specified by the ROHC working group. [ROHC] will from this point be referenced to as ROHC. These guidelines should simplify incorporation of the robust header compression algorithms into cellular system like those standardized by 3GPP, 3GPP2, ETSI, etc, and also into specific link layer protocols such as PPP. The document should also enable preparation of this incorporation without requiring detailed knowledge about the final header compression scheme. Relevant standardization groups standardizing link layers should, aided by this document, include required functionality in "their" link layers to support robust header compression. Hence, this document clarify the requirements [ROHC] put on lower layers, while the requirements on ROHC may be found in [REQ]. 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. Svanbro [Page 3] INTERNET-DRAFT Lower Layer Guidelines for Robust HC October 10, 2000 3. General guidelines 3.1 Error detection All current header compression schemes [RFC1144, RFC2507, RFC2508] rely on lower layers to detect errors in (compressed) headers. This is usually done with link layer checksums covering at least the compressed header. However, all error detecting mechanisms may fail to detect some bit errors and cause residual bit errors (undetected bit errors). Lower layers MUST provide error detection for at least ROHC headers. ROHC has been designed not to increase residual bit error rate (for reasonable residual error rates) compared to the case when no header compression is used. Headers passed up to the header decompressor should, however, have a residual bit error close to zero. It is RECOMMENDED that erroneous headers are passed up to the decompressor instead of being discarded before the decompressor, but in that case an indication that the header has errors MUST be included to the decompressor together with the erroneous header. 3.2 Inferred header field information Some fields of the RTP/UDP/IP headers may be classified as inferred, that is their values are to be inferred from other values or from an underlying link layer. An underlying link layer MUST therefore provide the header decompressor with the following information: Packet Length (IPv4) Information about the received packet (with the compressed header) length MUST be provided by the link layer. Payload Length (IPv6) Information about the received packet (with the compressed header) length MUST be provided by the link layer. Length (UDP) This field is redundant with the Packet Length (IPv4) or the Payload Length (IPv6) field. Hence, the value of this field MUST in some way be provided to the decompressor. In summary, all of the fields above relate to the length of the packet the compressed header is included in. All three fields may thus be inferred by the decompressor if one packet length value is signaled from the link layer to the decompressor on a per packet basis. This packet length value should be the length of the received packet including the (compressed) header. Svanbro [Page 4] INTERNET-DRAFT Lower Layer Guidelines for Robust HC October 10, 2000 3.3 Handling of header size variations It is desirable for many cellular link layer technologies that bit rate variations and thus packet size variations are minimized. However, there will always be some variation in compressed header sizes since there is a trade-off between header size variations and compression efficiency, and also due to events in the header flow and on the channel. Variations in header sizes cause variations in packet sizes depending on variations of payload size. The following will only treat header size variations caused by ROHC and not packet size variations due to variations of payload size. The link layer MUST in some manner support varying header sizes from 40 bytes (full RTP/UDP/IPv4 header) or 60 bytes (full RTP/UDP/IPv6) down to 1 byte for the minimal compressed header. It is likely that the small compressed headers dominate the flow of headers, and that the largest headers are sent rarely, e.g. only a few times in the initialization phase of the header compression scheme. Header size variations and thus packet size variations depend on numerous factors. Un-predictable changes in the RTP, UDP or IP headers may cause compressed headers to momentarily increase in size as well as header sizes may depend on packet loss rate at lower layers. Header size distributions depend also on the mode ROHC operate in. However, for e.g. a voice application, carried by RTP/UDP/IPv4, with a constant speech frame size and silence suppression, the following basic header size changes may be considered as typical: In the very beginning of the speech session, the ROHC scheme is initialized by sending full headers called IR/DYN. These are the largest headers and with sizes depending on basically IP-version. For IPv4 the size is approximately 40 bytes, and for IPv6 approximately 60 bytes. The IR/DYN headers are used typically during one round trip time, possible interleaved with compressed headers. After that, usually only compressed headers are sent. Compressed headers may vary in size from 1 byte up to several bytes. The smallest compressed headers is used when there is no unpredictable changes in header fields, typically during a talk spurt. In the beginning of a talk spurt, compressed header sizes may increase with one or a few bytes momentarily. Apart from increases due to new talk spurts, compressed headers may increase in size momentarily due to unpredictable changes in header fields. ROHC provides some means to limit the amount of produced header sizes. In some cases a larger header than needed may be used to limit the number of header sizes used. Padding octets may also be used to fill up to a desired size. Chapter 6.3 Implementation parameters in [ROHC] provide optional implementation parameters that make it possible to mandate how a ROHC implementation should operate, for instance to mandate how many header sizes that may be used. Svanbro [Page 5] INTERNET-DRAFT Lower Layer Guidelines for Robust HC October 10, 2000 3.4 Negotiation of header compression parameters ROHC have some parameters that needs to be set at an initial setup phase. Which header compression profile to use may have to be determined and also what kind of context identification (CID) mechanism to use. The lower layers supporting ROHC MUST include mechanisms for negotiating header compression parameters such as, CID usage and/or header compression profiles. It is RECOMMENDED that the lower layer have mechanisms that support re-negotiations of these parameters. 3.5 Demultiplexing of flows onto logically separated channels In some cellular technologies there are a demultiplexing of flows onto radio bearers suitable to the particular flows, i.e., onto logically separated channels. For instance, real-time flows such as voice and video may be carried on logically separated bearers. It is RECOMMENDED that this kind of demultiplexing is done in the lower layers supporting robust header compression. By doing so, the need for context identification in the header compression scheme is reduced. If there is a one to one mapping between flow and logical channel, there is no need at all for context identification at the header compression level. 3.6 Packet type identification Header compression schemes like [RFC2507, RFC2508] have relied on the underlying link layer to identify different kinds of headers by means of packet type identifiers on link layers. This kind of mechanism is not necessary for ROHC. The packet type identification is incorporated in the ROHC scheme and there is thus, no need for such a mechanism in the link layer. If ROHC is used together with header compression schemes requiring packet type identification at the link layer, e.g. [RFC2507, RFC2508], or if ROHC is used on top of link layers where packet type identifiers already are present, it is RECOMMENDED that one (1) ROHC packet type identifier is supported on lower layers. Thus, only one ROHC packet type is needed to mix ROHC and e.g. RFC2507 flows or to support ROHC on links where packet type identifiers already are present. 3.7 Packet duplication Exact duplications of one and the same packet may waste transmission resources and is in contradiction to compression. Even so, packet duplication may occur for various reasons. Packet duplication may also occur in different places along the path for a packet. ROHC can handle packet duplication before the compressor but it is RECOMMENDED that such packet duplications are avoided. Lower layers Svanbro [Page 6] INTERNET-DRAFT Lower Layer Guidelines for Robust HC October 10, 2000 MUST NOT duplicate packets on the path between ROHC compressor and decompressor. 3.8 Feedback packets ROHC consist of three modes; Unidirectional mode (U-mode), bidirectional optimistic mode (O-mode) and bidirectional reliable mode (R-mode). A brief description of the modes can be found in chapter 4.4 in [ROHC]. In U-mode it is not necessary to send any feedback from the decompressor to the compressor. O-mode and R-mode requires however that feedback messages from the decompressor to the compressor may be sent. Feedback message consist of small ROHC internal packets without any application payload. It is possible in ROHC to piggy-back feedback packets onto regular packets with ROHC compressed headers and payload, if there is ROHC type of compression in both the forward and reverse direction. However, this piggy-backing may not be desired or possible in some cases. Lower layer MUST support transport of feedback packets from decompressor to compressor if ROHC is to be used in O-mode or R-mode. Lower layers MUST support transport of small stand-alone feedback packets if piggybacking of feedback packets is not used. The feedback packets from the decompressor SHOULD be delivered as soon as possible to the compressor. 4. Cellular system specific guidelines An important group of link layer technologies where robust header compression will be needed are future cellular systems, which may have a very large number of users in some years. The need for header compression is large in these kinds of systems to achieve spectrum efficiency. Hence, it is important that future cellular systems can efficiently incorporate the robust header compression scheme. 4.1 Handover procedures One cellular specific property that may affect header compression is mobility and thus, handover (i.e., change of serving base station or radio network controller). The main characteristics of handovers relevant for robust header compression are: the length of the longest packet loss event due to handover (i.e. the number of consecutive packet losses); relocation of header compression context when necessary. Depending on the location of the header compressor/decompressor in the radio access network and the type of handover, handover may or may not cause disruptions or packet loss events in the (compressed) header flow relevant for the header compression scheme. For instance, Svanbro [Page 7] INTERNET-DRAFT Lower Layer Guidelines for Robust HC October 10, 2000 if soft handover is used and if the header compressor/decompressor reside above the combining point for soft handover, there will be no extra packet losses visible to the decompressor due to handover. In hard handovers, where packet loss events due to handover is introduced, the length of the longest consecutive packet loss is most relevant and should thus, be minimized. Handover SHOULD NOT cause significant long events of consecutive packet loss. Significant in this context relates to the kind of loss tolerable for the carried real-time application. If hard handovers are performed, which may cause significant long events of consecutive packet loss, it is RECOMMENDED that the radio access network notifies the compressor when such a handover has started and completed. The cellular system supporting robust header compression may also have internal mechanisms for transferring the header compression context between nodes where contexts may reside, at or before handover. If the cellular system does not have any internal mechanism for transferring header compression context between nodes, the context relocation may be done by the header compression scheme by means of a context refresh. The header compression scheme may perform a new header compression initialization, e.g. by sending full headers. This will, however, introduce an increase in the average header size dependent on how often a transfer of context is needed. In this latter case, the lower layers MUST indicate to the header compressor that such a handover has occurred, so that it knows when to refresh the context. Chapter 6.3 Implementation parameters in [ROHC] provide optional implementation parameters that make it possible to trigger e.g. a complete context refresh. 4.2 Unequal error detection Chapter 3.1 states that ROHC requires error detection from lower layers for at least the compressed header. However, some cellular technologies may differentiate the amount of error detection for different parts of a packet. For instance, it could be possible to have a stronger error detection for the header part of a packet, if the application payload part of the packet is less sensitive to errors, e.g. some cellular types of speech codecs. ROHC does not require UED from lower layers, ROHC requires only a error detection mechanism that detects errors in at least the header part of the packet. There is thus no requirement on lower layers to provide separate error detection for the header and payload part of a packet. However, overall performance may be increased if UED is used. Svanbro [Page 8] INTERNET-DRAFT Lower Layer Guidelines for Robust HC October 10, 2000 For example, if equal error detection is used in the form of one link layer checksum covering the entire packet including both header and payload part, any bit error will cause the packet to be discarded at the ROHC decompressor. It is not possible to distinguish between errors in the header and the payload part of the packet with this error detection mechanism and the ROHC decompressor must assume that the header is damaged, even if the bit error hit the payload part of the packet. If the header is assumed to be damaged, it is not possible to ensure correct decompression and that packet will thus be discarded. If the application is such that it tolerates some errors in the payload, it could have been better to deliver that packet to the application and let the application judge whether the payload was usable or not. Hence, with a unequal error detection scheme where it is possible to separate detection of errors in the header and payload part of a packet, more packets may be delivered to applications in some cases for the same lower layer error rates. The final benefit depend of course on the cost of UED for the radio interface and related protocols. 4.3 Unequal error protection Some cellular technologies can provide different error probabilities for different parts of a packet, unequal error protection (UEP). For instance, the lower layers may provide a stronger error protection for the header part of a packet compared to the payload part of the packet. ROHC does not require UEP. UEP may be beneficial in some cases to reduce the error rate in ROHC headers, but only if it is possible to distinguish between errors in header and payload parts of a packets. I.e. only if unequal error detection (UED) is used. The benefit of UEP depend of course on the cost of UEP for the radio interface and related protocols. 5. Author's Addresses Krister Svanbro Tel: +46 920 20 20 77 Box 920 Fax: +46 920 20 20 99 Ericsson Erisoft AB Email: krister.svanbro@ericsson.com SE-971 28 Lulea, Sweden Svanbro [Page 9] INTERNET-DRAFT Lower Layer Guidelines for Robust HC October 10, 2000 6. References [ROHC] Carsten Borman et al, "Robust Header Compression (ROHC)", Internet-Draft (work in progress), October 2000. [REQ] Mikael Degermark, "Requirements for robust IP/UDP/RTP header compression", Internet Draft (work in progress), June 2000. [RFC1144] Van Jacobson, "Compressing TCP/IP Headers for Low-Speed Serial Links", RFC 1144, February 1990. [RFC2507] Mikael Degermark, Bjorn Nordgren, Stephen Pink, "IP Header Compression", RFC 2507, February 1999. [RFC2508] Steven Casner, Van Jacobson, "Compressing IP/UDP/RTP Headers for Low-Speed Serial Links", RFC 2508, February 1999. [RFC2509] Mattias Engan, Steven Cassner, "IP Header Compression over PPP", RFC2509, February 1999 [RFC1661] Simpson, W., Ed., "The Point-To-Point Protocol (PPP)", STD 51, RFC 1661, July 1994. This Internet-Draft expires in April 2001. Svanbro [Page 10]