LSMA WG Minutes LSMA Chair Michael Myjak Co-chair/secretary, Jon Crowcroft WG E-mail list: lsma@gmu.edu lsma-request@gmu.edu Work in progress: draft-myjak-lsma-scenarios-01.txt draft-pullen-lsma-limitations-02.txt draft-ietf-lsma-requirements-01.txt 0. Agenda Bashing Agenda was agreed. Minutes of Last Meeting/AOB items: no comments on these beyond those already received on mail list and no AOB Attendees/Blue form was passed around and turned in. 1. Michael Myjak - LSMA Scenarios Draft Update Presentation [Slides available] Mike presented these - the main update was material on a broader variety of large scale applications categorization drawn from the presentation at the Reliable Multicast Research Group by Mark Handley. This extended the work well beyond the DIS/HLA into other (more typically Internet friendly, and recognizable in some cases) realms. Assumptions about definitions of "large", based on receiver set size, dynamics, and sender set size were discussed - it was noted that tree type and denseness are other relevant dimensions and their interpretation can vary by application. 2. Mark Pullen - Limitations of IETF Standards for LSMA Draft Update Presentation [Slides available] Mark presented (and distributed updated paper copies) of the very latest (as of 7/12/97) draft - additional material on aggregation of flows, inter-domain routing requirements, and a low latency naming service were discussed, in addition to the baseline document. Some key distinguishing features/points on DIS were pointed out. Mark's main point being that the proportion of a group that can be classified as senders is typically 90%, which is very, very different from most other LSMA categories. Defense-oriented security requirements (e.g. possibly stronger cryptographic properties and robust against traffic analysis) are also other attributes which place DIS/HLA in a separate category. Shortfalls specific in current IETF protocols, and work plans were also noted in discussion. There was a useful discussion of the problems in identifying quantitative (rather than just qualitative) protocol limitations - Mark suggested that it was difficult to estimate accurately a limitation until the technology had come within a small factor (often 2) of the limitations being addressed. 3. Peter Bagnell - update on BT LSMA Requirements Draft This needs more work which is ongoing. [Slides Available] Peter presented material on a middle-ware architecture for multimedia and adaptive multicast applications. - Jon Crowcroft asked some questions about the multicast scaling factors (e.g. feedback signals, layering approaches, granularity and time scales of layering), two level architectures, combining regional multicast and transcoder/filtering application level cache/proxy/server systems between regions.) Peter noted that their requirements work had identified 62 (or more) dimensions along which one could characterize these applications. He noted that they needed several levels of indirection (e.g. media type specific and so on), and their middle-ware architecture (similar to ActiveMovie/DirectShow) captured some of this functionality. Peter noted that an implementation would proceed from early 1998. 4. Christophe Diot (INRIA) on minimal multicast generic requirements for LSVE [Slides available] Christophe presented material on experiences with MiMaze over the Mbone with International tests with 15 users. Reliability was not the key problem, Christophe noted. Scaling and congestion control were key problems - he drew attention to the data rates implicit from the many sender approach in DIS, and discussed realist statistics for the current Mbone. Christophe then qualitatively outlined the way that utility varied with group size, and how this influenced the design - an extended example based on a three level membership in a football match (e.g. players interacting with each other and ball, rest of teams, and stadium attendees). He placed explicit requirements for hierarchical groups, trees and group control, as well as a low latency naming system to permit dynamic mappings and re-mappings (e.g. for HLA like ideas of grid and so on (extent range and so on) to be supported). 5. Mike McCarthy from NPS spoke on work led by Don Brutzman on the dis-java-vrml project. NPS is building a large-scale multicast application for widespread testing over the MBone. Mike noted that software details available at http://www.stl.nps.navy.mil/dis-java-vrml 6. Michael Myjak for Dr. Richard Weatherly on Data Distribution Management and the High Level Architecture. Unfortunately Dr. Weatherly couldn't make it, but Mike summarized nicely [Slides Available]. The HLA work, being lead by the U.S. Defense Modeling and Simulation Office, describes a multi-level extremely configurable "routing space" concept. This is used for mapping the path of relevant data from a source to a destination. The example given included geographic grids to multicast set of addresses, but other mappings are possible. These mappings need not be regular or evenly spaced/scaled to allow for different levels of detail and interest. This is similar to the LSVE work in various European projects (e.g. Coven, Diot, etc.) and is relevant many LSMAs. Experience from actual STOW deployment (4000 rather than 50,000 entities) had been fairly positive - modifications/enhancements to the routing space concepts would be carried forward. 7. Jon Crowcroft Attempt to quantify scaling issues of LSMA and DIS/HLA and current Internet Architecture. Jon noted that several existing and new working groups were being established to work issues raised by LSMA participation. For example, Flow aggregation work is being carried out now in Diff Serv (part of internet services (int-serv), currently but may branch out). Flow Signal aggregation - is new work (taken by IBM and ISI i-ds) in the RSVP group on signaling for aggregated flows. Scaling issues brought out in LSMA will be addressed. Low latency name/address management is a current working item in the Multicast Allocation (Malloc) BOF and will likely to become a new IETF WG very soon. Finally, Low latency, hierarchical group leave/join issues could be the next generation of IGMPv4 work - This would be in the Inter Domain Multicast Routing (IDMR) working group, and we should request this of the IDMR chairs. We also identified IPSEC (IP Security) requirements - special needs from LSMA especially from games and DIS (e.g. non cheating of timestamps as per MiMAZE and HIPPARCH ideas from INRIA, non traffic analysis of multicast as per Pullen notes) needs new IPSEC work. This is additional input to IPNG/v6 only, but needs to spawn IPSEC new key distribution ideas - some of these were discussed here and in the Multicast Netnews BOF. If this becomes a working group, it could act as a focus for at least part of the scalable key distribution for non-repudiation, privacy and prevention of denial of service and traffic analysis attacks presented previously. Some of the ideas from the BT work could be continued as contributions to the ADAPTS BOF's Information RFC - more general Middleware might be out of scope for the IETF in general since it did not really require Internet Protocols per se. Scaling work on Inter-domain multicast has finally started to gel with BGMP in IDMR - however, this does not do address group aggregation, only forwarding state - this needs attention. This appears to be somewhere between IDMR and Malloc. IGMP v4 work to use this would have to be in IDMR, logically. The MiMaze experience could be related to real costs by looking at the pricing in commercial ISP's "Cast" services (e.g. UUNET UU-Cast service) which are typically 1-many, and protect low bandwidth sites from 'pulling down' traffic onto the NAPs (or whatever UUNET calls them!). A hybrid model of server based and 1-many (e.g. see expressway work by Cheriton and Holbrook at Stanford) might be more tractable for today's limited ISPs. Still remaining is the issue of many-many with large percentages of the senders also acting as receivers. Key success in terms of resource reservation (key requirement for many LSMA applications, though not all) is the new differentiated services work items taking place in int-serv. This really looks like they can deliver an effective performance to large numbers of participants however, it remains to be seen how these would apply to systems with _MANY_ participants in a dense sending group. Blake/Elleson's presentions at int-serv showed how hard it is to dimension a diff-serv profile for unpredictable spatial traffic distributions, but this is NOT unpredictable, so it might represent a class that ISPs could understand (whereas Diot's more dynamic lower cost one actually is much harder to deal with since it does use dynamics to keep the total traffic cross section down. Unfortunately it does so at the expense of locality. RTCP removal for some types of games is an item for the AVT WG. 2-level systems with a name service are actually becoming common in other networks that are not currently part of the commercial Internet. (e.g. ISIS and TIBCO support a subject based/content based address publish subscriber service, with application level server/filter gateways, connected by IP multicast. This currently doesn't fly on the commercial Internet, but could easily be the eventually outcome of the combined work suggested here.) Some of the LAN Adaptor card limitations could be bought up at the IEEE in the relevant 802 sub-committees. 8. Working Group Final Call on i-ds, to go to IESG for review as Informational RFCs Mark's i-d is ready to go - the other two need modest work, and then can be submitted for last call on the list, and thence hopefully to the IESG by February, or even before. 9. New WGs Jon Crowcroft asked if we believed we needed an explicit follow up WG (or indeed needed to ask for a charter extension/update for this one) - no-one required this - the working group will close its work on the list with the submission of the RFCs. 10. AOB and Wrap up This was the last meeting. Thanks to all contributors and see you in other WGs at the LA IETF.