CURRENT MEETING REPORT
Reported by Cyndi Mills, BBN Planet
Minutes of the Realtime Traffic Flow Measurement BOF (RTFM)
Fifty-seven people attended the BOF, including participants
representing views from service providers, implementors and consumers
of Internet services.
Peter Lothberg put the service provider view, explaining what he
wants from a Flow Measurement system. This included the
measurement of traffic patterns, performance data, bottleneck data,
changes in user behavior due to backpressure, etc.
Nevil Brownlee gave a summary of the existing Internet Drafts
(architecture and MIB) and an overview of NetraMet, a flow
measurement implementation with meters running on several platforms
ranging from low-end PCs under DOS to workstations under Unix. Nevil
outlined how NeTraMet is being used in New Zealand, Australia and
elsewhere for network analysis and planning as well as live network
troubleshooting, and talked about operational experience and uses.
Sig Handelman presented IBM Research's implementation of the MIB
and plans for extending this to the measurement of IBM Global
Network and server performance.
Bob Moskowitz outlined the Automotive Industry Action Group's desire
that they be able to independently use tools such as these to measure
the network performance and management capabilities of Internet
service providers in order to qualify them for participation in virtual
private networks running over public network facilities.
The initial RTFM proposal was discussed in some detail. It differs from
other probe efforts in that instead of trying to collect a mass of data and
post-analyze it, data reduction is done at the point of measurement.
Interesting (or uninteresting) flows can be aggregated as "summary
flows." For example, a set of 1,700,000 flows might be summarized
within the meter as follows:
o 200,000 packets in flows of type "uninteresting,"
o 1,400,000 packets in flows of type "special interest a,"
o 20 packets in specified flow b, for each flow fulfulling characteric c, a
dynamically allocated ennumerated list of flows, e.g. 10 packets in a
flow with the following value of c, 12 packets in a separate flow
with another value of c, etc.
The ability to do this reduces the data at an early stage with a very
small level of effort. Counting can be done in parallel with a packet-
forwarding decision at a level of effort which is roughly comparable to
forwarding calculations. The meter may be in a server, standalone
(probe), or in a switching device.
The BOF reached consensus that a Traffic Measurement working group
should be formed in the Operations Area to standardise traffic
measurement. Traffic measurement requirements and concerns will be
posted to the mailing list:
[email protected] (send a subscribe
message to rtfm-request...).
A charter for the proposed working group was discussed. The resulting
draft charter is as follows:
Draft Charter: Real Time Traffic Flow Measurement (RTFM)
Co-chairs:
Nevil Brownlee, The University of Auckland
Barry Greene, Singapore Telecom
Sig Handelman, IBM Research
o Review existing work in traffic flow measurement, including that of
the RMON and Internet Accounting working groups and published
work from independent researchers.
o Produce an improved Traffic Flow Model considering at least the
following:
- efficient hardware implementation
- effect of IPv6 on traffic measurement
- extension of the accounting model to widen the range of measurable
quantities
- simpler ways to specify flows of interest
- maintain existing focus on data reduction capabilities
o Develop the Flow Meter MIB as a 'standards track' document with
the IETF.
Goals and Milestones:
February 1996 Resubmit set of revised
documents to RFC Editor for
publication as experimental
RFCs. 1) Traffic Flow
Measurement Background:
revised document based on
RFC1272; 2) Traffic Flow
Measurement Architecture
Document: revised document
based on draft-brownlee-acct-
arch-report-02.txt; 3) Traffic
Flow Meter MIB: revised
document based on draft-
brownlee-acct-meter-mib-
03.txt; and, 4) Traffic Flow
Examples: new document
recording experience with
Meter MIB, number three
above.
Circulate list of references to
earlier work to mailing list for
information and discussion.
Solicit opinions and requests for
additional quantities to be
measured, ways to measure
them, and issues for future
high-speed networks.
March 1996 Produce outline for 'New
Traffic Flow Model' document
July 1996 Publish 'New Traffic Flow
Model' document as Internet
Draft and begin work on
�implementation� document for
new traffic flow model
Novembewr 1996 Submit 'New Traffic Flow
Model' to IESG as RFC and
publish 'implementation'
document as Internet Draft
