Editor's note: These minutes have not been edited.
Minutes of LSMA WG/BOF(II) Meeting in San Jose, Dec 9
Jon Crowcroft, UCL,
[email protected]
Mike Myjak, Florida,
[email protected]
WG E-mail list:
[email protected]
[email protected]
0 Introduction
Mike Myjak introduced the BOF, and announced that it had now been
approved as a Working Group, under two area directors, Harald
Alvestra (Applications AD) and Allyn Romanow (Transport Area Co-Chair
with Allison Mankin) would be responsible for this groups output.
Mike went through the groups charter, and introduced a couple of
minor modifications for Harald approval (to appear after the WG) -
he also mentioned the various WGs that we would interact with.
Harald explained that this was an unusual WG in that its output would
be a set of requirements on other IETF WGs, in particular, areas where
there were shortcomings, and a critique of the protocols from those
other groups.
Later Allyn outlined how the Transport Area Directors believed that
there is a requirement for a whole new work item on reliable
multicast, and that there would be a session on this as part of the
Open Area Meeting on wednesday.
Other WG's we might interact with include (not just), IDMR, Mbone
Deployment, MMUSIC, AVT, QOSR, etc etc.
We then had a series of presentations:
1. Steve Seidensticker (
[email protected])
presented his paper on "scenarios"
(draft-myjak-lsma-scenarios-00.txt)
This concentrated on the communications patterns requirements from
the DIS and SISO scenarios, including various stages (entity state,
emissions, weather, fire control,) and styles (peer-peer, client
server, multicast n the full dynamic, many-to-many sense), and talked
about convergence of DIS and VRML work....mentioning IEEE 1278.1 and
DIS 2.04.
[See Snowcrash, a fictional work by the UK Sci Fi novelist, Neal
Stephenson for more details).
2. Mark Pullen
[email protected]
(draft-pullen-lsma-limitations-00.txt)
Mark concentrated on detail performance requirements, such as group
size, group dynamics, traffic rates, join/leave latency requirements,
data transmission latency and jitter requirements and so on.
He talked about the HLA model and the requirement to have filtering
(at 10kbps per sender, if a receiver was in all groups, they would
receive 1Gbps for a group size of 100k participants, not atypical for
a DIS STOW scenario).
Transport is mainly UDP, since its mainly "real time" (mbone like) -
they have occasional, but important reliable delivery requirements,
which are too low latency to make TCP an option (and involve too many
arbitrary participants to permit a mesh of pre-configured TCP
connections 100k**2).
Q1 from WH audience - what about symmetry of links or paths?
jon answered "outside our scope, below our level"
mark answered "so long as latency requirement is met, not our
problem".
Q2. What about SRM for reliability
jon said "yes" but see Rel Mul work...
or Area Directors Internet Draft on this for other references
on reliable multicast transport.
Q3. What about balance between routing, and bridging
to help manage complexity, scaling of traffic.
Good point....
Harald pointed out that there was a new requirement: "Given size of
environment, what is the cost of an end system to participate?"
3. Ken Carlberg (UCL/SAIC) presented a paper on scaling sport for DIS
and associated IETF problems.
This talk mainly covered multicasting and resource reservation
problems, especially in the context of IP on ATM - areas such as
aggregation of tree state, reservation state, and multipath multicast
routing were mentioned as requirements to meet the higher level
performance requirements that Pullen had outlined.
4.Rajesh Talpade (GATECH) discussed his work on
Single Connection Emulation, an approach to providing multi-way
operation for applications that have been built to communicate via a
single TCP connection (legacy retro-fitting).
Q., What about the T.20 work on t.mbftp
yes, thats a similar approach
Q. what about ack aggregation?
yes, that too is an alternate approach
5. Don Brotzman (Naval Postgraduate School)
presented VRTP, an architecture for a transfer protocol to sit along
side HTTP to transfer VRML in the same way as the Web transfers HTML
Rather than being just a peer-peer protocol, or a client server
protocol, VRTP would be a framework for a spectrum of systems that
include both.
Don mentioned the VRML symposium to be held in California in Febuary,
and stated his goal, to build the CBone (Cyberspace Backbone) by 2
years from now.
6. Christophe Diot (
[email protected]) (INRIA) presented the INRIA work on the imaze, a
distributed gaming environment, very similar to the DIS, with some neat
ideas on synchronisation, and so forth.
The WG meeting wound up, and a lot of people wanted more details...
6
--------------
Slides will be on the archive site, along with the LSMA mail
archive....TBD!
