Network Working Group H. Hannu, Ericsson INTERNET-DRAFT J. Christoffersson, Ericsson Expires:
NovemberDecember 2002 S. Forsgren K. Leung Z. Liu, Nokia R. Price, Siemens/Roke Manor MayJune 03, 2002 SigComp - Extended Operations <draft-ietf-rohc-sigcomp-extended-03.txt><draft-ietf-rohc-sigcomp-extended-04.txt> 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, firstname.lastname@example.org. Abstract This document describes mechanismshow to be used with Signaling Compression (SigComp),implement certain mechanisms in SigComp (Signaling Compression), RFC XXX, thatwhich can significantly improve the compression efficiency compared to using per-messagesimple per- message compression. The mechanisms, such as explicit acknowledgementsSigComp uses a UDVM (Universal Decompressor Virtual Machine) for decompression, and shared compressionthe mechanisms described in this document are possible to implement using the UDVM instructions defined and explainedin this document.RFC XXX. Table of contents 1. Introduction....................................................2 2. Terminology.....................................................2 3. Architectural View of Feedback..................................4 4. State Reference Model...........................................4 5. Extended Mechanisms.............................................6 6. Implications on SigComp........................................11 7. Security Considerations........................................15 8. IANA Considerations............................................15 9. Acknowledgements...............................................15 10. Authors' Addresses.............................................15 11. Intellectual Property Right Considerations.....................16 12. References.....................................................16 1. Introduction This document describes how to implement mechanisms with [SIGCOMP] to significantly improve the compression efficiency compared to per- messages compression. One such mechanism is to use previously sent messages in the SigComp compression process, referred to as dynamic compression. In order to utilize information from previously sent messages, it is necessary for a compressor to gain knowledge about the reception of these messages. For a reliable transport, such as TCP, this is guaranteed. For an unreliable transport however, the SigComp protocol can be used to provide such a functionality itself. That functionality is described in this document and is referred to as explicit acknowledgements. Another mechanism that will improve the compression efficiency of SigComp, especially when SigComp is applied to protocols that are of request/response type, is shared compression. This involves using received messages in the SigComp compression process. In particular the compression of the first few messages will gain from shared compression. Shared compression is described in this document. For better understanding of this draft the reader should be familiar with the concept of [SIGCOMP]. 2. Terminology The reader should consult [SIGCOMP] for definitions of terminology, since this draft uses the same terminology. Further terminology is defined below. Compressor Entity that encodes application messages using a certain compression algorithm and keeps track of state that can be used for compression. The compressor is responsible for ensuring that the messages it generates can be decompressed by the remote UDVM. Decompressor The decompressor is responsible for converting a SigComp message into uncompressed data. Decompression functionality is provided by the UDVM. Dynamic compression Compression relative to messages sent prior to the current compressed message. Explicit acknowledgement Acknowledgement for a state thatstate. The acknowledgment is explicitly sent from a decompressor to its remote compressor. The acknowledgement should be piggybacked onto a SigComp message in order not to create additional security risks. Shared compression Compression relative to messages received by the local endpoint prior to the current compressed message. Shared state A state used for shared compression consists only of an uncompressed message. This makes the state independent of compression algorithm. State identifier Reference used to access a previously created item of state. - shared_state_id State identifier of a shared state. - acked_state_id State identifier of a state that is acknowledged as successfully saved by the decompressor. 3. Architectural View of Feedback SigComp has a request/response mechanism to provide feedback between endpoints, see Figure 1. This particular functionality of SigComp is used in this document to provide support for the mechanisms described in this document. +--------------------+ +--------------------+ | Endpoint 1 | | Endpoint 2 | | +--------------+ | | +--------------+ | | | Compressor 1 | | | |Decompressor 2| | | | [------------+--+--------------+--+--] * | | | +-|-------^----+ | | +--|---|-------+ | | | | | | | | | | | | | | | | | | | | | | | | | | +-|-------|----+ | | +--v---|-------+ | | | * [----+--+--------------+--+------] | | | |Decompressor 1| | | | Compressor 2 | | | +--------------+ | | +--------------+ | +--------------------+ +--------------------+ Figure 1. Architectural view The feedback functionality of SigComp is used in this document to provide a mechanism for a SigComp endpoint to confirm which states have been established by its remote SigComp endpoint during the lifetime of a SigComp compartment. The established state confirmations are referred to as acknowledgments. Depending on the established states this particular type of feedback may or may not be used to increase the compression efficiency. The following sections describe how the SigComp functionality of providing feedback information is used to support the mechanisms described in this document. Section 4 describes the state reference model of SigComp. Section 5 continues with a general description of the mechanisms and Section 6 describes the implications of some of the mechanisms on basic SigComp. 4. State Reference Model A UDVM may want to save the status of its memory, and this status is referred to as a state. As explained in [SIGCOMP] a state save request may or may not be granted by the application. For later reference to a saved state, e.g. if the UDVM is to be loaded with this state, a reference is needed to locate the specific state. This reference is called a state identifier. 4.1. Overview of State Reference with Dynamic Compression When compressor 1 compresses a message m it uses the information corresponding to a SigComp state that its remote decompressor 2 has established and acknowledged. If compressor 1 wishes to use the new state for compression of later messages it must save the new state. The new state contains information from the former state and from m. When an acknowledgement is received for this new state, compressor 1 can utilize the new state in the compression process. Below is an overview of the model together with an example of a message flow. Saved state(s) A state which is expected to be used for compression/decompression of later messages. Acked state(s) An acked state is a saved state for which the compressor has received an acknowledgement, i.e. the state has been established at the remote decompressor. The compressor must only use states that are established at the remote decompressor, otherwise a decompression failure will occur. For this reason, acknowledgements are necessary, at least for unreliable transport. Compressor 1 Decompressor 2 +---+ +---+ | C | | D | +---+ +---+ Saved Acked | | Saved State(s) State(s) | | State(s) -----------------------+------------+------------------ s0 s0 | | s0 s1=s0+m1 | --m1(s0)-->| | <--ack(s1) | s0,s1 s0,s1 s0,s1 | | | | s0,s1 s0,s1 | --m2(s1)-->| (m2 Lost) s2=s1+m1 | | | | s0-s2 s0,s1 | | s3=s1+m3 | --m3(s1)-->| s0,s1 | | | | | <--ack(s3) | s0,s1,s3=s1+m3 s0-s3 s0,s1,s3 | | Figure 2. Example of message flow. 5. Extended Mechanisms The following subsections give a general description of the extended mechanisms. 5.1. Explicit Acknowledgement Scheme For a compressor to be able to utilize a certain state it must know that the remote decompressor has access to this state. In the case where compressed messages can be lost or misordered on the path between compressor and decompressor, an acknowledgement scheme must be used to notify the remote compressor that a certain state has been established. Explicit acknowledgements can be initiated either by UDVM-code uploaded to the decompressor by the remote compressor or by the endpoint where the states have been established. These two cases will be explained in more detail in the following two sections. 5.1.1. Remote Compressor Initiated Acknowledgements This is the case when e.g. compressor 1 has uploaded UDVM bytecode to decompressor 2. The UDVM bytecode will use the requested feedback field in the announcement information and the returned feedback field in the SigComp header to obtain knowledge about established states at endpoint 2. Consider Figure 3. An event flow for successful use of remote compressor initiated acknowledgements can be as follows: (1): Compressor 1 saves e.g. state(A). (2): The UDVM bytecode to initiate a state save for state(A) is either carried in the compressed message, or can be retrieved by decompressor 2 from a state already saved at endpoint 2. (3): As compressor 1 is the initiator of this acknowledgement it can use an arbitrary identifier to be returned to indicate that state(A) has been established. The identifier needs to consist of enough bits to avoid acknowledgement of wrong state. To avoid padding of the feedback items and for simplicity a minimum of 1 octet should be used for the identifier. The identifier is placed at the location of the requested_feedback_item [SigComp]. The END-MESSAGE instruction is used to indicate the location of the requested_feedback_item to the state handler. (4): The requested feedback data is now called returned feedback data as it is placed into the SigComp message at compressor 2. (5): The returned feedback item is carried in the SigComp message according to Figure 4: see Section 6.1 and [SIGCOMP]. (6): The returned feedback item is handled according to: Section 7 of [SIGCOMP] +--------------+ (2) +--------------+ | Compressor 1 |--------------------------->|Decompressor 2| +------^-------+ +-------^------+ | (1) (3) | +---v---+ +---v---+ |State | |State | |handler| |handler| +---^---+ +---^---+ | (6) (4) | +------v-------+ (5) +-------v------+ |Decompressor 1|<---------------------------| Compressor 2 | +--------------+ +--------------+ Figure 3. Simplified SigComp endpoints 5.1.2. Local Endpoint Initiated Acknowledgements When explicit acknowledgements are provided by an endpoint, the SigComp message will also carry acknowledgements, so-called acked_state_id: see Section 2. Consider Figure 3, an event flow for successful use of explicit endpoint initiated acknowledgements can be as follows: (1): Compressor 1 saves e.g. state(A). (2): The UDVM bytecode to initiate a state save for state(A) is either carried in the compressed message, or can be retrieved by decompressor 2 from a state already saved at endpoint 2. (3): A save state request for state(A) is passed to the state handler using the END-MESSAGE instruction. The application may then grant the state handler permission to save state(A): see [SIGCOMP]. (4): Endpoint 2 decides to acknowledge state(A) to endpoint 1. The state identifier (acked_state_id) for state(A) is placed in the SigComp message sent from compressor 2 to decompressor 1. (5): The UDVM bytecode to initiate (pass) the explicit acknowledgement to endpoint 1 is either carried in the compressed message, or can be retrieved by decompressor 1 from a state already saved at endpoint 1. (6): The acked_state_id for state(A) is passed to the state handler by placing the acked_state_id at the location of the "returned SigComp parameters" [SIGCOMP], which location is given to the state handler using the END-MESSAGE instruction. Note: When the requested feedback length is non-zero endpoint initiated acknowledgements should not be used, due to possible waste of bandwidth. When deciding to implement this mechanism one should consider whether this is worth the effort as all SigComp implementations will support the feedback mechanism and thus have the possibility to implement the mechanism of Section 5.1.1. 5.2. Shared Compression To make use of shared compression a compressing endpoint saves the uncompressed version of the compressed message as a state (shared state). As described in Chapter 2 the references to a shared state is referred to as shared_state_id. The shared state's parameters state_address and state_instruction must be set to zero. The state_retention_priority must be set to 65535, and the other state parameters are set accordingly to [SIGCOMP]. This is because different compression algorithms may be used to compress application messages traveling in different directions. The shared state is also created on a per-compartment basis, i.e. the shared state is stored in the same memory as the states created by the particular remote compressor. The choice of how to divide the state memory between "ordinary" states and shared states is an implementation decision at the compressor. Note that new shared state items must not be created unless the compressor has made enough state memory available (as decompression failure could occur if the shared state pushed existing state out of the state memory buffer). A compressing endpoint must also indicate to the remote compressor that the shared state is available, but only if the local decompressor can retrieve the shared state. The retrieval of the shared state is done accordingly to the state retrieval instruction of the UDVM. Consider Figure 3. An event flow for successful use of shared compression can be as follows: (1): Compressor 1 saves e.g. state(M), which is the uncompressed version of the current application message to be compressed and sent. (2): The UDVM bytecode to indicate the presence of state(M) at endpoint 1 is either carried in the compressed message, or can be retrieved by decompressor 2 from a state already saved at endpoint 2. (3): The SHA-1 instruction is used at endpoint 2 to calculate the shared_state_id for state(M). The indication is passed to the state handler, by placing the shared identifier at the location of the "returned SigComp parameters" [SIGCOMP]. The location of the "returned SigComp parameters" is given to the state handler using the END-MESSAGE instruction. (4): If endpoint 2 uses shared compression, it compares the state identifier values in the "returned SigComp parameters" information with the value it has calculated for the current decompressed message received from endpoint 1. If there is a match then endpoint 2 uses the shared state together with the state it would normally use if shared compression is not supported to compress the next message. (5): The UDVM bytecode that will use the shared state (state(M)) in the decompression process at decompressor 1 is either carried in the compressed message, or can be retrieved by decompressor 1 from a state already saved at endpoint 1. 5.3. Maintaining State Data Across Application Sessions Usually, signaling protocols (e.g. SIP) employ the concept of sessions. However, from the compression point of view, the messages sent by the same source contain redundancies beyond the session boundary. Consequently, it is natural to maintain the state data from the same source across sessions so that high performance can be achieved and maintained, with the overhead amortized over a much longer period of time than one application session. Maintaining states across application sessions can be achieved simply by making the lifetime of a compartment longer than the time duration of a single application session. Note that the states here are referring to those stored on a per-compartment basis, not the locally available states that are stored on a global basis (i.e. not compartment specific). 5.4. Use of User-Specific Dictionary The concept of the user-specific dictionary is based on the observation that for protocols such as SIP, a given user/device combination will produce some messages containing fields that are always populated with the same data. Take SIP as an example. Capabilities of the SIP endpoints are communicated during session initiation, and tend not to change unless the capabilities of the device change. Similarly, user- specific information such as the user's URL, name, and e-mail address likely will not change on a frequent basis, and will appear regularly in SIP signaling exchanges involving a specific user. Therefore, a SigComp compressor could include the user-specific dictionary as part of the initial messages to the decompressor, even before any time critical signaling messages are generated from a particular application. This enables increased compression efficiency once the messages start to flow. Obviously, the user-specific dictionary is a state item that would be good to have as a cross-session state: see Section 5.3. 5.5. Checkpoint State The following mechanism can be used to avoid decompression failure due to reference to a non-existent state. This may occur in three cases: a) a state is not established at remote SigComp endpoint due to loss of a SigComp message; b) a state is not established due to insufficient memory; c) a state has been established but was deleted later due to insufficient memory. When a compressor sends a SigComp message that will create a new state on the decompressor side, it can indicate that the newly created state will be a checkpoint state by setting state_retention_priority [SIGCOMP] to the highest value sent by the same compressor. In addition, a checkpoint state must be explicitly acknowledged by the receiving decompressor to the sending compressor. Consider Figure 3. An event flow for this kind of state management can be as follows: (1): Compressor 1 saves e.g. state(A), which it would like to have as a checkpoint state at decompressor 2. (2): The UDVM bytecode to indicate the state priority ([SIGCOMP] - state_retention_priority) of state(A) and initiate a state save for state(A) is either carried in the compressed message, or can be retrieved by decompressor 2 from a state already saved at endpoint 2. (3): A save state request for state(A) is passed to the state handler using the END-MESSAGE instruction, including the indication of the state priority. The application grants the saving of state(A): see [SIGCOMP]. (4): An acknowledgement for state(A) (the checkpoint state) is returned to endpoint B using one of the mechanisms described in Section 5.1. Note: To avoid using a state that has been deleted due to insufficient memory a compressor must keep track of the memory available for saving states at the remote endpoint. The SigComp parameter state_memory_size which is announced by the SigComp feedback mechanism can be used to infer if a previous checkpoint state has been deleted (by a later checkpoint state creation request) due to lack of memory. 5.6. Implicit Deletion for Dictionary Update Usually a state consists of two parts: UDVM bytecode and dictionary. When dynamic compression is applied, new content needs to be added to the dictionary. To keep an upper bound of the memory consumption such as in the case for a low end mobile terminal, existing content of the dictionary must be deleted to make room for the new content. Instead of explicitly signaling which parts of the dictionary need to be deleted on a per message basis, an implicit deletion approach may be applied. Specifically, some parts of the dictionary are chosen to be deleted according to a well-defined algorithm that is known and applied in the same way at both compressor and decompressor. For instance, the algorithm can be part of the predefined UDVM bytecode that is agreed between the two SigComp endpoints. As input intoto the algorithm, one provides the total number of bytes to be deleted. The algorithm then specifies which parts of dictionary are to be deleted. Since the same algorithm is applied at both SigComp endpoints, there is no need for explicit signaling on a per message basis. This may lead to higher compression efficiency due to the avoidance of signaling overhead. It also means more robustness as there are no signaling bit on the wire that are subject to possible transmission errors/losses. 6. Implications on SigComp The extended features will have implications on the SigComp messages sent between the compressor and its remote decompressor, and on how to interpret e.g. returned SigComp parameters [SIGCOMP]. However, except for the mandatory bytes of the SigComp messages [SIGCOMP], the final message formats used are implementation issues. Note that an implementation that does not make use of explicit acknowledgements and/or shared compression is not affected, even if it receives this kind of feedback. 6.1. Implications on SigComp Messages To support the extended features, SigComp messages must carry the indications and information addressed in Section 5. For example to support shared compression and explicit acknowledgements the SigComp messages need to convey the following information: - The acked_state_id as described in Sections 2 and 5.1. - The shared_state_id as described in Sections 2 and 5.2. Figure 4, depicts the format of a SigComp message according to: [SIGCOMP] 0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7 +---+---+---+---+---+---+---+---+ +---+---+---+---+---+---+---+---+ | 1 1 1 1 1 | T | len | | 1 1 1 1 1 | T | 0 | +---+---+---+---+---+---+---+---+ +---+---+---+---+---+---+---+---+ | | | | : returned feedback item : : returned feedback item : | | | | +---+---+---+---+---+---+---+---+ +---+---+---+---+---+---+---+---+ | | | code_len | : partial state identifier : +---+---+---+---+---+---+---+---+ | | | code_len | destination | +---+---+---+---+---+---+---+---+ +---+---+---+---+---+---+---+---+ | | | | : remaining SigComp message : : uploaded UDVM bytecode : | | | | +---+---+---+---+---+---+---+---+ +---+---+---+---+---+---+---+---+ | | : remaining SigComp message : | | +---+---+---+---+---+---+---+---+ Figure 4. Format of a SigComp message. The format of the field "remaining SigComp message" is an implementation decision by the compressor which supplies the UDVM bytecode. Therefore there is no need to specify a message format to carry the information necessary for the extended features described in this document. Figure 5, depicts an example of what the "remaining SigComp message" with support for shared compression and explicit acknowledgements, could look like. Note that this is only an example; the format is an implementation decision. 0 1 2 3 4 5 6 7 +---+---+---+---+---+---+---+---+ | Format according to Figure 4 | : except for the field called : | "remaining SigComp message" | "remaining SigComp message" field +---+---+---+---+---+---+---+---+ -------- | s | a | r | Reserved | | +---+---+---+---+---+---+---+---+ | | | | : shared_state_id* : Present if 's' is set | | | +---+---+---+---+---+---+---+---+ | | | | : acked_state_id* : Present if 'a' is set | | | +---+---+---+---+---+---+---+---+ | | | | : Rest of the SigComp message : | | | v +---+---+---+---+---+---+---+---+ -------------- Figure 5. Example of SigComp message for some of the extended features. 'r' : If set, then a state corresponding to the decompressed version of this compressed message (shared state) was saved at the compressor. * : The length of the shared_state_id and acked_state_id fields are of the same length as the partial state identifier. 6.2. Extended SigComp Announcement/Feedback Format This section describes how the "returned_SigComp_parameters" [SIGCOMP] information is interpreted to provide feedback according to Section 5.1 and 5.2. The partial_state_identifiers correspond to the hash_value for states that have been established at the remote endpoint after received SigComp messages, i.e. these are acknowledgements for established states and may be used for compression. The partial_state_identifiers may also announce "global state" that is not mapped to any particular compartment and is not established upon the receipt of a SigComp message. It is up to the implementation to deduce what kind of state each partial_state_identifier refers to, e.g. an acknowledged state or a shared state. In case a SigComp message that includes state identifiers for shared states and/or acknowledged states is received by a basic SigComp implementation, these identifiers will be ignored. The I-bit of the requested feedback format is provided to switch off the list of locally available state items. An endpoint that wishes to receive shared_state_id must not set the I-bit to 1. The endpoint storing shared states and sending the list of locally available states to its remote endpoint must be careful when taking the decision whether to excluded or include different types of the locally available states (i.e. shared states or states of e.g. well- known algorithms) from/to the list. 6.3. Acknowledgement Optimization If shared compression is used between two endpoints (see Figure 1) then there exists an optimization, which if implemented makes an acked_state_id in the SigComp message unnecessary: Compressor 1 saves a shared state(M), which is the uncompressed version of the current compressed message (message m) to be sent. Compressor 1 also sets bit 'r' (see Figure 5), to signal that state(M) can be used by endpoint 2 in the compression process. The acked_state_id for state(S), which was created at endpoint 2 upon the decompression of message m, may not have to be explicitly placed in the compressed messages from compressor 2 if the shared state(M) is used in the compression process. When endpoint 1 notices that shared state(M) is requested by decompressor 1, it implicitly knows that state(S) was created at endpoint 2. This follows since: * Compressor 1 has instructed decompressor 2 to save state(S). * The indication of shared state(M) would never have been received by compressor 2 if state(S) had not been successfully saved, because if a state save request is denied then the corresponding announcement information is discarded by the state handler. Note: Endpoint 1's state handler must maintain a mapping between state(M) and state(S) for this optimization to work. Note: The only state acknowledge by this feature acknowledges is the state that was created by combining the state used for compression of the message and message itself. For any other case the acked_state_id has to be used. Note: There is a possibility that state(S) is discarded due to lack of state memory even though the announcement information is successfully forwarded. This possibility must be taken into account (otherwise a decompression failure may occur), and can be so by using the SigComp parameter state_memory_size which is announced by the SigComp feedback mechanism. The endpoint can use this parameter to infer if a state creation request has failed due to lack of memory. 7. Security Considerations The features in this document are believed not to add any security risks to the ones mentioned in [SIGCOMP]. 8. IANA Considerations This document does not require any IANA involvement. 9. Acknowledgements Thanks to Carsten Bormann, Christopher Clanton, Miguel Garcia, Lars-Erik Jonsson, Khiem Le, Mats Nordberg, Jonathan Rosenberg and Krister Svanbro for valuable input. 10. Authors' Addresses Hans Hannu Box 920 Ericsson AB SE-971 28 Lulea, Sweden Phone: +46 920 20 21 84 E-Mail: email@example.com Jan Christoffersson Box 920 Ericsson AB SE-971 28 Lulea, Sweden Phone: +46 920 20 28 40 E-Mail: firstname.lastname@example.org Stefan Forsgren E-Mail: email@example.com Ka Cheong Leung E-mail: firstname.lastname@example.org Zhigang Liu Nokia Research Center 6000 Connection Drive Irving, TX 75039, USA Phone: +1 972 894-5935 E-Mail: email@example.com Richard Price Roke Manor Research Ltd Romsey, Hants, SO51 0ZN, United Kingdom Phone: +44 1794 833681 E-mail: firstname.lastname@example.org 11. Intellectual Property Right Considerations The IETF has been notified of intellectual property rights claimed in regard to some or all of the specification contained in this document. For more information consult the online list of claimed rights. 12. References [SIP] M. Handley et. al., SIP: Session Initiation Protocol, RFC 2543, Internet Engineering Task Force, March 1999 [SIGCOMP] R. Price et. al., Signaling Compression (SigComp), Internet Draft (work in progress), MayJune 2002. <draft-ietf-rohc-sigcomp-06.txt><draft-ietf-rohc-sigcomp-07.txt> This Internet-Draft expires NovemberDecember 03, 2002.