CURRENT_MEETING_REPORT_
Reported by Ken Schroder/BBN
CIP Minutes
Agenda
Status Reports
o ST-II
o COIP-K
o FP
o MCHIP
Collaboration Plans
o Research, experiments
Meeting Report
The Connection Oriented Internetwork Protocol Working Group (CIP) is
developing a set of protocols and resource management algorithms to
support guaranteed service, packet switched communication in an
internet. Applications in the areas of wide area video conferencing and
distributed simulation would both benefit from service guarantees.
Elements of this support include resource reservation, flow regulation,
instrumentation and enforcement mechanisms to ensure acceptable
bandwidth, end-to-end delay and delay variation. Approaches for
allowing reservations to be renegotiated as the workload changes are
also anticipated.
Claudio Topolcic, Working Group Chair, opened the meeting. The goal of
this meeting was to review what had been accomplished since the
Vancouver meeting and to plan what will be done during the next three
months. We were particularly interested in understanding how the work
each group member was doing might compliment one other.
RFC-1190 ``Experimental Internet Stream Protocol, Version 2 (ST-II)''
has been released. ST-II is an IP-layer protocol that provides
1
end-to-end service guarantees across an internet. It was designed
through earlier efforts of the Working Group to replace the Internet
Stream Protocol originally defined in IEN-119.
ST-II implementation status was presented by Ken Schroder. Portions of
the control protocol are currently operating at BBN on an Ethernet.
They expect to:
o Pass data application to application over Ethernet by the end of
December.
o Integrate T1 support by end of January.
The protocol implementation is expected to be operating in the DARPA
sponsored DARTNET in February. Support will include connection setup
and tear down, hop identifier negotiation, and add/delete targets.
ST-II will then be used as a protocol testbed for exploring
instrumentation and algorithms that:
o Ensure proper priority traffic handling to ensure that time
guarantees are met.
o Provide predictable estimates of delay and delay variance.
o Guarantee that network switching elements meet end-to-end
performance promised to applications.
o Enforce that application traffic cannot exceed the resources level
it originally requested.
The issuing of RFC-1190 signaled the end, at least for now, of the ST
track that this Working Group was following. The Working Group will
continue to study connection oriented protocols.
FP Flow Protocol work was presented by Lixia Zhang. They are using IP
option fields to implement the flow protocol. This approach has
simplified the work required and allows the protocol to coexist with IP,
since standard gateways will forward the packets. Developing a
customized protocol would not have offered those benefits. The current
implementation goals include support for:
o Lixia's Flow Protocol
o Fair queuing algorithms
o Timestamp ordered driver queues to support priority scheduling
2
They have plans to experiment with dynamic rate adjustment, including
selectively throttling traffic sources (rather than all sources) to
handle congestion control. They hope to make TCP use FP in the future.
They cited several difficulties they encountered with the current
approach.
o Clock granularity is too coarse for traffic generator applications
programs to use for generating packets at specific rates
o Table lookup inefficient: hard to get small universal identifiers
o Fair Queuing for IP is implemented on a per TCP connection basis.
The current implementation uses source and destination host IP
addresses plus port numbers as the connection identifier.
Performance measurement was discussed. They timestamp packets at
source, destination and all intermediate routers. Since transmission
and propagation delays are known, queuing delay can be calculated.
Potential future work includes:
o Virtual clock testing. The virtual clock was implemented but not
tested because queues don't build up on Sparcs with Ethernet.
(Ethernet is much faster.)
o FP providing reliability by selective retransmission
o Host pacing
FP/ST sharing was discussed. It was felt that some of the enforcement
mechanism supported by the virtual clock Lixia's flow protocol could be
integrated into the ST-II network layer. This would require integration
of the timestamp ordering mechanisms and supplying various flow
parameters. The potential for more extensive integration will be
discussed after the ST-II implementation is working.
Resource management work at Berkeley was presented by Hui Zhang. Their
work includes explicit delay and jitter control. Packets are marked
with the desired transmission time and buffered until the deadline
arrives. This works to limit jitter. Studies they have performed
suggests this will also reduce the buffer space requirements of the
overall network.
Connection Oriented IP Kernel was presented by Guru Parulkar. The
COIP-K is meant to provide a core set of functions--application and
network interface, data forwarding and state machine
3
management--expected to be needed by high performance protocols such as
ST-II. Their goal is to provide a reusable foundation in which resource
management protocol research can be performed more easily.
o Chuck Cranor will return to work on software shortly
o They expect to have it debugged in January
o Can implement resource enforcement, potentially by incorporating
Lixia's virtual clock code
There was some discussion about the availability and suitability of
COIP-K to the ST-II and FP efforts. We plan to revisit this in January
after initial implementation is available.
MCHIP was presented by Guru Parulkar. This is a connection oriented
resource management protocol that Guru has been working on. There are
three basic elements:
1. Resource requirements characterized by peak rate, average rate and
burstiness.
2. Perpetual Congrams (PiCons) are routed using reservations and
virtual circuits, e.g., through ATM networks.
3. Server can provide resource allocations for unmanaged datagram
networks, e.g., Ethernet. (There was some dispute as to whether
this was doable in the general case, whether source routing would
provide an adequate solution, and how much constraints would have
to be relaxed for it to work.)
The meeting concluded after discussions of what next steps to take. The
potential combining of COIP-K, ST-II, and FP into a single COIP will be
explored in January. Many elements of FP resource management and
enforcement seem complimentary and compatible with the ST-II
implementation, which provides connection setup and management
facilities. The COIP-K is intended to be compatible with these and
other protocols.
We plan to meeting, ideally by video conference, in late January to
discuss how more of our work can be integrated. At that point, working
versions of COIP-K and ST-II should both be available.
Attendees
Ashok Agrawala agrawala@cs.umd.edu
Robert Braden braden@isi.edu
4
Kevin Fall kfall@ucsd.edu
Gurudatta Parulkar guru@flora.wustl.edu
Ken Schroder schroder@bbn.com
Claudio Topolcic topolcic@bbn.com
Hui Zhang hzhang@tenet.berkeley.edu
Lixia Zhang lixia@parc.xerox.com
5