CURRENT_MEETING_REPORT_
Reported by Andy Nicholson/Cray Research, Inc.
CBNR BOF Minutes
Description:
We held a BOF on this subject at the 20th IETF in St. Louis, and this
is a continuation of that interchange. However, the only attendees from
the St. Louis meeting present at this meeting were the Cray Research,
Inc. representatives.
While working with circuit-switched T3 networks, developers at Cray
Research, Inc., determined that there would be advantages to defining a
standard way to control certain classes of network resources through the
internet. In the case of a circuit-switched T3 line, the line should be
switched on only when there are active transport connections which can
fully utilize the service. Due to the high cost of the resource,
under-utilization would be particularly undesirable. The developers
believe that this capability might have other applications in the
internet and that an effort should be made to define a standard
protocol.
Minutes:
Due to the small size and informality of the meeting, no formal Minutes
were taken. This record is believed to be reasonably accurate and
proper credit given to the originators of the ideas and concepts
presented. Andy Nicholson, who is preparing this report, apologizes for
any errors or omissions.
A variety of new issues were brought up at this meeting, and it was
encouraging to note that there were as many non-Cray people as Cray
employees. Most of the discussion centered around the concrete example
of the Circuit-Switched T3 services being prototyped by Cray Research,
Inc.
The first issue raised centered on local routing to the T3 adapter.
This would include routing to any controlled device. The prototypes
assume that a particular network link will be used for transfer of data
packets, thus static routing is implied. There is concern that this
perspective may lead to the use of static routing between the requesting
host and the controlled resource. There was general agreement that this
should not happen.
Another issue concerned recursive conditioning of resources. A host in
control of a link might need to pass a request to another host through
the controlled link so that further downstream links may be conditioned.
This should be possible.
Fred noted that some comparisons could be made with regard to this
capability. For example, this is similar to the switching which takes
1
place through the telco fabric as calls are routed. There is also a
similarity to X.25. For example, TP0 will create a link over x.25 when
a connection is established.
Matt did not think that this could be a widely deployed function;
however, Fred noted that this might be useful in any kind of
fundamentally switched service, i.e. ISDN or mobile hosts. This seems
like something that is in the future.
In the Cray Research, Inc. prototype, most of the support code is in
the kernel. Matt and Fred were concerned that perhaps this should all
reside in user space. This leads to two fundamentally different
approaches. For everything to be in user space, either special commands
must be executed to condition the network before running applications,
or the applications must make special library calls. If everything is
in the kernel, then these services can be transparent to users and
applications.
These discussions led us to a very different conclusion from the last
meeting. All agreed that I would finish an informational RFC relating
our experiences with the switched T3 services in time for review by the
community before the next IETF. If possible, I will also document the
protocol we are using.
At the 22nd IETF we will once again hold a BOF to gauge interest in
these facilities. At that meeting we will determine whether to continue
any work through the IETF.
Attendees
Fred Baker fbaker@emerald.acc.com
David Borman dab@cray.com
Matt Mathis mathis@psc.edu
Andy Nicholson droid@cray.com
John Seligson johns@ultra.com
Jeff Young jsy@cray.com
2