INTERNET DRAFT P Bagnall Large-scale Multicast Applications Working Group R Briscoe Expiration: 21 May 1998 A Poppitt BT 21 Nov 1997 Taxonomy of Communication Requirements for Large-scale Multicast Applications draft-ietf-lsma-requirements-01.txt 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 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.'' To learn the current status of any Internet-Draft, please check the ``1id-abstracts.txt'' listing contained in the Internet- Drafts Shadow Directories on ftp.is.co.za (Africa), nic.nordu.net (Europe), munnari.oz.au (Pacific Rim), ds.internic.net (US East Coast), or ftp.isi.edu (US West Coast). Abstract The intention of this draft is to define a classification system for the communication requirements of any large-scale multicast application (LSMA). It is very unlikely one protocol can achieve a compromise between the diverse requirements of all the parties involved in any LSMA. It is therefore necessary to understand the worst-case scenarios in order to minimise the range of protocols needed. Dynamic protocol adaptation is likely to be necessary which will require logic to map particular combinations of requirements to particular mechanisms. Standardising the way that applications define their requirements is a necessary step towards this. Classification is a first step towards standardisation. 1. Introduction This taxonomy consists of a large number of parameters that are considered useful for description of communication requirements of LSMAs. To describe a particular application, each parameter would be assigned a value. Typical ranges of values are given wherever possible. Failing this, the type of any possible values is given. The parameters are collected into ten or so higher level categories, but this is purely for convenience. The parameters are pitched at a level considered meaningful to application programmers. However, they describe communications not applications - the terms "3D virtual world", or "shared TV" might imply communications requirements, but they don't accurately describe them. Assumptions about the likely mechanism to achieve each requirement are avoided where possible. The exception to this is that receiver initiated join to multicast address groups [refmcast] on an open access Internet is assumed. While the parameters describe communications, it will be noticed that few requirements concerning routing etc. are apparent. This is because applications have few direct requirements on these second order aspects of communications. Requirements in these areas will have to be inferred from application requirements (e.g. latency). The taxonomy is likely to be useful in a number of ways: - most simply, it can be used as a checklist to create a requirements statement for a particular LSMA. Example applications will be classified [bagnall97] using the taxonomy in order to exercise (and improve) it - because strictest requirement have been defined for many parameters, it will be possible to identify worst case scenarios for the design of protocols - because the scope of each parameter has been defined (per session, per receiver etc.), it will be possible to highlight where heterogeneity is going to be most marked - a step towards standardisation of the way LSMAs define their communications requirements. This could lead to standard APIs between applications and protocol adaptation middleware - identification of limitations in current Internet technology for LSMAs to be added to the LSMA limitations draft [limitations] - identification of gaps in Internet Engineering Task Force (IETF) working group coverage This approach is intended to complement that used where application scenarios for Distributed Interactive Simulation (DIS) are proposed [scenarios] in order to generate network design metrics (values of communications parameters). Instead of creating the communications parameters from the applications, we try to imagine applications that might be enabled by stretching communications parameters. The above introduction assumes all the items under the "Further Work" section (near the end) have been completed. As they haven't, the reader is advised to read that section next! 2. Definitions The following terms have no agreed definition, so they will be defined for this document. Session a happening or gathering consisting of flows of information related by a common description that persists for a non-trivial time (more than a few seconds) such that the participants (be they humans or applications) are involved and interested at intermediate times may be defined recursively as a super-set of other sessions Secure session a session with restricted access A session or secure session may be a sub and/or super set of a multicast group. A session can simultaneously be both a sub and a super-set of a multicast group by spanning a number of groups while time-sharing each group with other sessions. 2.1. Definitions of Roles Defining all possible roles is not possible. The roles in a communication are application dependant. 3. Taxonomy 3.1 Summary of Communications Parameters Before the communications parameters are defined, typed and given worst- case values, they are simply listed for convenience. Also for convenience they are collected under classification headings. Reliability packet loss Transactional Guaranteed Tolerated loss Semantic loss component reliability setup fail-over time mean time between failures fail over time during a stream Ordering Ordering type Timeliness Synchronicity Burstiness Jitter expiry latency optimum bandwidth tolerable bandwidth required by time and tolerance host performance fair delay frame size content size Session Control initiation start time end time duration active time session burstiness atomic join late join allowed ? temporary leave allowed ? late join with catch-up allowed ? potential streams per session active streams per sessions Session Topology # of senders # of receivers Directory fail-over timeout (see Reliability: fail-over time) mobility Security authentication strengh tamper-proofing non-repudiation strength denial of service action restriction privacy retransmit prevention strength membership criteria membership principals collusion prevention fairness action on compromise Security dynamics mean time between compromises compromise detection time limit compromise recovery time limit Payment & Charging for what charge basis content services when who pays whom prevention of onward re-sale Costing of communications elements of cost cost epochs quotations charging costs 3.2 Definitions, types and strictest requirements The terms used in the above table are now defined for the context of this document. Under each definition, the type of their value is given and where possible worst-case values and example applications that would exhibit this requirement. There is no mention of whether a communication is a stream or a discrete interaction. An attempt to use this distinction as a way of characterising communications proved to be remarkably unhelpful and was dropped. 3.2.1 Reliability 3.2.1.1 Reliability - Packet loss Transactional ------------- When multiple operations must occur atomically, transactional communications guarantee that either all occur or none occur and a failure is flagged. Type: Boolean - on/off Strictest requirement - on Example application: bank credit transfer, debit and credit must be atomic. NB: Transactions are potentially much more complex, but it is believed this is an application layer problem. Guaranteed ---------- Guarantees communications will succeed in certain cases. Type: enumerated Deferrable if communication fails it will be deferred until a time when it will be successful. Guaranteed the communication will succeed so long as all necessary components are working. No guarantee - failure will not be reported. Strictest requirement - deferred Example application: stock quote feed guaranteed NB: the application will need to set parameters to more fully define Guarantees, which the middleware may translate into, for example, queue lengths. Tolerated loss -------------- This specifies the proportion of data from a communication that can be lost before the application becomes completely unusable. Type: fraction Strictest requirement: 0% Example application: video 40% Semantic loss ------------- The application specifies how many and which parts of the communication can be discarded if necessary. type: identifiers - names disposable app level frames strictest requirement - no loss allowed example application: video feed - P frames may be lost, I frames not. 3.2.1.2. Component Reliability Setup Fail-over time -------------------- The time before a failure is detected and a replacement component is invoked. This is not directly an application requirement. Type: time Strictest Requirement: application dependent Example application: Name lookup - 5 seconds Mean time between failures -------------------------- Type: time Strictest requirement: indefinite Example application: xxx Fail over time during a stream ------------------------------ The time between a stream breaking and a replacement being set up. Type: time Strictest Requirement: latency Example application: xxx 3.2.2. Ordering Ordering type ------------- Specifies what ordering must be preserved for the application Type: boolean trueidempotent false> enumeration timing values: global per sender none sequenced values: global per sender none causal values: global per sender none Strictest requirement - global timed, sequenced & causal Example application : Game - global causal (to make sure being hit by bullet occurs after shot is fired!) 3.2.3. Timeliness There is a meta-requirement on timeliness. If hard real-time is required then the interpretation of all the other requirements changes. Failures to achieve the required timeliness must be reported before the communication is made. By contrast soft real-time means that there is no guarantee that an event will occur in time. However statistical measures can be used to indicate the probability of completion in the required time, and policies such as making sure the probability is 95% or better could be used. Hard-real time: Boolean - hard/soft Synchronicity ------------- To make sure that separate elements of a session are correctly synchronised with respect to each other Type: time Strictest requirement 80ms Example application: TV lip-sync value 80ms Burstiness ---------- This is a measure of the variance of bandwidth requirements over time. Type: fraction - variation in b/w as fraction of b/w for variable b/w communications fraction - duty cycle (fraction of time at peak b/w) for intermittent b/w communications. Strictest requirement: variation -> max b/w, duty cycle -> 0 Example application: sharing video clips, with chat channel - sudden bursts as clips are swapped. Compressed Audio - difference between silence and talking NB: More detailed analysis of communication flow (eg max rate of b/w change or Fourier Transform of the b/w requirement) is possible but as complexity increases usefulness and computability decrease. Jitter ------ Jitter is a measure of variance in the time taken for communications to traverse from the sender (application) to the receiver, as seen from the application layer. Type: time Strictest requirement - <1ms Example application: audio streaming - <1ms NB: A jitter requirement implies that the communication is a real-time stream. Expiry ------ This specifies how long the information being transferred remains valid for. Type: date (seconds since...) Strictest requirement - for ever Example application: key distribution - 3600 seconds (valid for at least one hour) Latency ------- Time between initiation and occurrence of an action from application perspective. Type: time Strictest requirement application dependent Example application - audio conference 20ms NB: where an action consists of several distinct sequential parts the latency budget must be split over those parts. For process control the requirement may take any value. Optimum Bandwidth ----------------- Bandwidth required to complete communication in time Type: bandwidth Strictest requirement - xxx Example application - I phone 8kb/s Tolerable Bandwidth ------------------- Minimum bandwidth that application can tolerate Type: bandwidth Strictest requirement - xxx Example application - I phone 4kb/s Required by time and tolerance ------------------------------ Time communication should complete by and time when failure to complete renders communication useless (therefore abort). Type: date - preferred complete time date - essential complete time Strictest requirement application dependent Example application: email - 1min & 1 day NB: bandwidth * duration size; only two of these parameters may be specified. An API though could allow application authors to think in terms of any two. Host performance ---------------- Ability of host to create/consume communication Type: application benchmark Strictest requirement: full consumption Example application: video - consume 15 frames a second NB: host performance is complex since load, media type, media quality, h/w assistance, and encoding scheme all affect the processing load. These are difficult to predict prior to a communication starting. To some extent these will need to be measured and modified as the communication proceeds. Fair delay ---------- Time between receipt of communication and response by the client should determine winner of race conditions, not the first response at the server. The alternative is that the transport should make sure that delivery is withheld until all reciepients have the data. The specified requirement determines what delay is acceptable between the first receiver getting the data and the last receiver getting the data (assuming no system failures, but including packet loss). Requirement: the variance in delay between users that is acceptable Type: time Strictest requirement: 10ms Example application: auction room - <10ms Frame size ---------- Size of logical data packets from application perspective Type: data size Strictest requirement: 6bytes (gaming) Example application: video data size of single frame update Content size ------------ The total size of the content (not relevant for continuous media) Type: data size Strictest requirement: N/A Example application: xxx 3.2.4. Session Control initiation ---------- which initiation mechanism will be used type: enumeration values : announcement invitation directive example application: corporate s/w update - directive start time ---------- time sender start sending! type: date strictest requirement: now example app: FTP - at 3am end time -------- type: date strictest requirement: now example app: FTP - now+30mins duration -------- (end time) - (start time) (duration), therefore only two of three should be specified. type: time strictest requirement: -> 0ms for discrete, indefinite for streams example app: audio feed - 60mins active time ------------ total time session is active, not including breaks type: time example app: spectator sport transmission session burstiness ------------------ expected level of burstiness of the session type: fixed point. variance as fraction of max bandwidth strictest requirement: bandwidth example app: commentary & slide show: 90% of max atomic join ----------- session fails unless a certain proportion of the potential participants accept an invitation to join. Alternatively, may be specified as a specific numeric quorum. type: fixed point (proportion required) or int (quorum) strictest requirement: 1.0 (proportion) example app: price list update, committee meeting Note: whether certain participants are essential is application dependent. late join allowed ? ------------------- does joining a session after it starts make sense type: Boolean & indirection strictest requirement: allowed example application: game - not allowed, indirect to spectator channel temporary leave allowed ? ------------------------- does leaving and then coming back make sense for session type: Boolean strictest requirement: allowed example application: FTP - not allowed late join with catch-up allowed ? --------------------------------- is there a mechanism for a late joiner to see what they've missed type: Boolean & indirection strictest requirement: allowed example app: sports event broadcast, allowed, indirect to highlights channel potential streams per session ----------------------------- total number of streams that are part of session, whether being consumed or not type: int strictest requirement: indefinite example app: football match mcast - multiple camera's, commentary, 15 streams active streams per sessions (ie max app can handle) --------------------------- maximum number of streams that an application can consume simultaneously type: int strictest requirements: indefinite example app: football match mcast - 6, one main video, four user selected, one audio commentary 3.2.5. Session Topology Note: topology may be dynamic. One of the challenges in designing adaptive protocol frameworks is to predict the topology before the first join. # of senders ------------ the number of senders is a result the middleware may pass up to the application type: int strictest requirement: indefinite example app: network MUD - 100 # of receivers -------------- the number of receivers is a results the middleware may pass up to the application type: int strictest requirement: indefinite example app: video mcast - 100,000 3.2.6. Directory fail-over timeout (see Reliability: fail-over time) ----------------- mobility -------- defines restrictions on when directory entries may be changed type: enumeration values: while entry is in use while entry in unused never strictest requirement: while entry is in use example app: voice over mobile phone, while entry is in use (as phone gets new address when changing cell). 3.2.7. Security The strength of any security arrangement can be stated as the expected cost of mounting a successful attack. This allows mechanisms such as physical isolation to be considered alongside encryption mechanisms. An example type would be 1970 UD$ (to inflation proof). Security is an othogonal requirement. Many requirements can have a security requirement on them which mandates that the cost of causing the system to fail to meet that requirement is more than the specified ammount. In terms of impact on other requirements though, security does potentially have a large impact so when a system is trying to determine which mechanisms to use and whether the requirements can be met security will clearly be a major influence. Authentication Strength ----------------------- Authentication aims to ensure that a principal is who they claim to be. For each role in a communication (see 2.1) there is a strength for the authentication of the principle who has taken on that role. The principal could be a person, organisation or other legal entity. It could not be a process since a process has no legal representation. Requirement: That the cost of hijacking a role is in excess of the specified amount. Each role is a different requirement. Type: Inflation proofed currency (eg 1970 UD$) Example application: inter-governmental conference Strictest Requirement: >budget of largest attacker Tamper-proofing --------------- This allows the application to specify how much security will be applied to ensuring that a communication is not tampered with. This is specified as the minimum cost of successfully tampering with the communication. Each non-security requirement has a tamper-proofing requirement attached to it. Requirement: The cost of tampering with the communication is in excess of the specified amount. Type: Inflation proofed currency: data is unchanged and complete? Inflation proofed currency: no replay of transmission is possible? Inflation proofed currency: data timeliness is assured (no malicious packet delay)? Strictest requirement: Each >budget of largest attacker Example application: stock price feed Pete - done Non-repudiation strength ------------------------ The non-repdiation strength defines how much care is taken to make sure there is a reliable audit trail on all interactions. It is measured as the cost of faking an audit trail, and therefore being able to "prove" an untrue event. There are a number of possible parameters of the event that need to be proved. The following list is not exclusive but shows the typical set of requirements. 1. Time 2. Ordering (when relative to other events) 3. Whom 4. What (the event itself) There are a number of events that need to be provable. 1. sender proved sent 2. receiver proves received 3. sender proves received. Type: inflation proofed currency Strictest requirement: >budget of largest attacker Example application: Full audit trail: billing based on usage logs. Random partial records: to deter users from fraud with the threat of the possibility of being able to detect it. Pete - done Denial of service ----------------- There may be a requirement for some systems (999,911,112 emergency sevices access for example) that denial of service attacks cannot be launched. While this is difficult (maybe impossible) in many systems at the moment it is still a requirement, just one that can't be met. Requirement: cost of launching a denial of service attack is greater than specified amount. Strictest Requirement: >budget of largest attacker Example application: web hosting, to prevent individual hackers stalling system. Pete - done Action restriction ------------------ For any given comunication there are a two actions, send and receive. Operations like adding to members to a group are done as a send to the membership list. Examining the list is a request to and receive from the list. Other actions can be generalised to send and receive on some communication, or are application level not comms level issues. Type: membership list/rule for each action. Note that both send and receive may use the same membership policy. Strictest requirement: send and receive have different policies. Example application: tv broadcast, sender policy defines transmitter, receiver policy is null. Pete - done Privacy ------- Privacy defines how well obscured a principals identity is. This could be for any interaction. A list of participants may be obscured, a sender may obscure their identity when they send. For each possible action there is a need to define the privacy required. There are also different types of privacy. For example knowing two messages were sent by the same person breaks the strongest type of privacy even if the identity of that sender is still unknown. For each "level" of privacy there is a cost associated with violating it. The requirement is that this cost is excessive for the attacker. Requirement: Level of privacy, expected cost to violate privacy level for openly identified anonymously identified (messages from the same sender can be linked) unadvertised (but tracable) undetectable Type: currency Strictest requirement: all levels >budget of attacker Retransmit prevention strength ------------------------------ This is extremely hard at the moment. This is not to say it's not a requirement. Requirement: the cost of retransmitting a secure piece of information should exceed the specified amount. Type:currency Pete - done Membership Criteria ------------------- If a principal attempts to participate in a communication then a check will be made to see if it is allowed to do so. The requirement is that certain principals will be allowed, and others excluded. Given the application is being protected from network details there are only two types of specification available, per user, and per organisation (where an organisation may contain other organisations, and each user may be a member of multiple organisations). Requirement: membership is limited to specified group, or specified group is excluded. Type: Boolean - include or exclude List of 1) users 2) organisations 3) user properties 4) org properties 5) hosts 6) hosts properties Pete Membership Principals --------------------- Entities that may join a rule-based secure session atomically. That is, a group of individuals is a principal if they can only all join or leave together. Principals can be considered as the SUBJECT field of an access control list, but this is not intended to imply ACL is a good method to use. Type: enumeration values: certified individuals certified group ids (corporations, organisations) lists (i.e. lists of lists, such as multicast groups, secure sessions) hosts Strictest requirement: mixture of all types. Example application: N/A Pete Collusion prevention -------------------- Which aspects of collusion it is required to prevent. Collusion is defined as malicious co-operation between members of a secure session. Superficially, it would appear that collusion is not a relevant threat in a multicast, because everyone has the same information, however, wherever there is differentiation, it can be exploited. Type: Boolean: time race collusion (true if needs preventing) Boolean: key encryption key (KEK) sharing Boolean: sharing of differential QoS (not strictly collusion as across sessions not within one) Strictest requirement: All true. Time race collusion is the most difficult one to prevent. Example application: A race where delay of the start signal may be allowed for, but one participant may fake packet delay while receiving the start signal from another participant. Pete - done Fairness -------- Fairness is orthogonal to many other requirements. Of particular interest are Reliability and Timeliness requirements. When a communication is first created the creator may wish to specify a set of requirements for these parameters. Principals which join later may wish to set tigher limits. Fairness enforces a policy that any improvement is requirement by one principal must be matched by all others, in effect requirements can only be set for the whole group. This increases the likelyhood that requirements of this kind will fail to be met. If fairness if not an issue then some parts of the network can use more friendly methods to achieve those simpler requirements. Requirement: that the variance of performance with respect to any other requirement is less than the specified amount. Type: delta of the requirement that needs to be fair. Example application: Networked game, latency to receive positions of players must be within 5ms for all players. Pete Action on compromise -------------------- The action to take on detection of compromise (until security reassured). Not sure this has anything to do with communications, really. Type: binary enumeration (Boolean?) values: warn but continue pause Strictest requirement: pause Example application: Secure video conference - if intruder alert, everyone is warned, but they can continue while knowing not to discuss sensitive matters (cf. catering staff during a meeting). Pete 3.2.7.1. Security Dynamics -------------------------- Security dynamics are the temporal properties of the security mechanisms that are deployed. They may affect other requirements such as latency or simply be a reflection of the security limitations of the system. The requirements are often concerned with abnormal circumstances (eg. system violation). Pete mean time between compromises ----------------------------- This is not the same as the strength of a system. A fairly weak system may have a very long time between compromises because it is not worth breaking in to, or it is only worth it for very few people. Mean time between compromises is a combination of strength, incentive and scale. Type: time Strictest requirement: xxx Example application: xxx Pete Compromise detection time limit ------------------------------- The average time it must take to detect a compromise (one predicted in the design of the detection system, that is). Type: time Strictest requirement: xxx Example application: xxx Pete Compromise recovery time limit ------------------------------ The maximum time it must take to re-seal the security after a breach. This combined with the compromise detection time limit defines how long the system must remain inactive to avoid more security breaches. For example if a compromise is detected in one minute, and recovery takes five, then one minute of traffic is now insecure and the members of the communication must remain silent for four minutes after detection while security is re-established. Type: time Strictest requirement: 1 second [NSA] Example application: xxx Pete 3.2.8. Payment & Charging This whole section is probably too far outside the scope of the LSMA working group and is unfinished anyway. Pete charging for what? ------------------ content services content distribution service QoS transmission security services directory services Type: Strictest requirement Example application: Pete Charge basis: content --------------------- (often different granularity to ownership basis) ownership of content own use/consumption unlimited limited (e.g. n copies or n "views") royalty-based (pay per "view") resale unlimited limited (e.g. n copies) royalty-based (could hit multicast enabled routers hard!) own use and resale own use and not resale resale and not own use Type: Strictest requirement Example application: Pete Charge basis: services ---------------------- subscription time expired in perpetuity pay per consumption time pay per "object" hybrids of all these Type: Strictest requirement Example application: Pete Payment: When? --------------- pre-paid deposit pre-pay (before any of each charge basis listed above) post-pay (before any of each charge basis listed above) periodically billed by usage free for introductory period credit and debit decoupled within tolerance time limited money limited time/money hybrid (formula, e.g. 30days if Pete Who pays whom? -------------- (directly as opposed to one collecting for another - every role player could charge directly, but these are the more likely ones to do so) This bit would be easier to read in two dimensions - who charges and who pays Media distributor (typically pays media owner (typically pays media advertiser)) Session owner (typically pays session advertiser (and might well pay media distributor)) Network providers (may be paid by session owner, but difficult as diff. receivers use diff.networks) Terminal owners (e.g. library, kiosk, arcade, pub etc) Matchmaker (in auctions, session directories etc.) advertising (to partially allay costs, or cover totally) sender pays (e.g. propaganda that receivers are paid to read!) pay to receive, paid to send (e.g. to encourage contribution of home videos - yuk!) Type: Strictest requirement Example application: Pete prevention of onward re-sale ---------------------------- Type: Strictest requirement Example application: Pete 3.2.8.1. Costing of communications See also topology Pete Elements of cost ---------------- terminal resources & QoS network resources & QoS server resources & QoS Type: Strictest requirement Example application: Pete Cost epochs ----------- up front investment fixed running costs independent of use variable (use-dependent) costs Type: Strictest requirement Example application: Pete quotations ---------- response time off-line "well known" pricing expiry spot pricing Type: Strictest requirement Example application: Pete Costs of charging ----------------- communications storage processing debt recovery fraud detection customer service (proving charges are valid) Type: Strictest requirement Example application: Pete 4. Mapping of Requirements to IETF Working Groups TBA Pete 5. Further Work Attempt to simplify! Refine definitions and types. In particular clarify where enumerations aren't intended to be "one of" types. Complete specifying worst case values & example apps. Identification of scope of each parameter (per session, per receiver, per sender etc.) to highlight potential heterogeneity problems Mapping between requirements and IETF Working Groups Exercising the taxonomy with some scenarios Exercising the taxonomy with some media-types which represent large sub- sets of application capabilities so can potentially be "macros" or shorthand to set values (or ranges) for a large number of parameters at once. 6. Security Considerations See comprehensive security section of taxonomy. References [Bagnall97] Bagnall Peter, Poppitt Alan, Example LSMA classifications [TBA] [refmcast] IP multicast ref [limitations] Pullen M, Myjak M, Bouwens C, Limitations of Internet Protocol Suite for Distributed Simulation in the Large Multicast Environment, Internet Draft, 26 Mar 1997, draft-ietf-lsma-limitations- 01.txt [scenarios] Seidensticker S, Smith W, Myjak W, Scenarios and Appropriate Protocols for Distributed Interactive Simulation, Internet Draft, 21 Jul 1997, draft-ietf-lsma-scenarios-01.txt [rmodp] Open Distributed Processing Reference Model (RM-ODP), ISO/IEC 10746-1 to 10746-4 or ITU-T (formerly CCITT) X.901 to X.904. Jan 1995. Catalogue entries: [blaze95] Blaze, Diffie, Rivest, Schneier, Shimomura, Thompson and Wiener, Paper on minimal key lengths for security in secret key ciphers? late 1995 [NSA] Wallner D, Harder E, Agee R, Key Management for Multicast: Issues and Architectures, National Security Agency, 1 July '97. Internet Draft draft-wallner-key-arch-00.txt 8. Authors' Addresses Bob Briscoe B54/74 BT Labs Martlesham Heath Ipswich, IP5 3RE England Phone: +44 1473 645196 Fax: +44 1473 640929 EMail: briscorj@boat.bt.com Home page: http://www.labs.bt.com/people/briscorj/ Peter Bagnall B54/74 BT Labs Martlesham Heath Ipswich, IP5 3RE England Phone: +44 1473 647372 Fax: +44 1473 640929 EMail: pbagnall@jungle.bt.co.uk Home page: http://www.labs.bt.com/people/bagnalpm/ Alan Poppitt B54/74 BT Labs Martlesham Heath Ipswich, IP5 3RE England Phone: +44 1473 640889 Fax: +44 1473 640929 EMail: apoppitt@jungle.bt.co.uk Home page: http://www.labs.bt.com/people/poppitag/