CURRENT_MEETING_REPORT_

Reported by J. Nevil Brownlee/University of Auckland and
Henry Clark/OARnet

Minutes of the Operational Statistics Working Group (OPSTAT)


Summary

The group decided on a large number of small improvements to the OPSTAT
statistics file format, and agreed to begin work on a new document,
``Statististics for Network People,'' which will set out recommendations
on effective performance measures for various types of network links.
An updated set of goals and milestones was also agreed upon.


Revision of RFC 1404

Nevil Brownlee had circulated an initial draft of a modified BNF for the
OPSTATS file.  Corrections to this and other changes from the version in
RFC 1404 were considered.  Most of these were accepted as trivial, but
the following provoked considerable discussion:


   o <router-name> should be <ASCII-string> to allow for arbitrary
     names.

   o Colon, comma and semicolon should be allowed as field separators.

   o <file-name> in <external-label-section> should be moved forward
     so that it follows BEGIN_LABEL.

   o It is not clear how the <tag-class> field is to be used.

   o Within the data section there must be a one-to-one mapping of tag
     variables to data values.

   o The <bw-sort> field is unnecessary.  Instead we should use bps as
     the only units, and allow floating-point values for <bw-value>.  A
     value of zero should mean `bandwidth unknown.'

   o Time resolution below one second would be useful.  We should allow
     floating-point values for <second>.

   o Should we attempt to provide for having separate variables of a
     sample stored in separate files?  The present format stores them in
     rows (one row per <data-field>).  Such a change was not considered
     to be worthwhile.

   o Should the file contain summary information about the quantity of
     data it contains?  This might reduce the amount of work a
     statistics server has to do to decide on the size of a given set of
     data, but would not help in deciding whether such a set is
     complete.  A statistics server implementation server might cache
     such information, but it does not seem worth trying to include this
     in the grammar.


Nevil will circulate a new draft of the grammar with these changes
incorporated.  Once these have been commented on, a new revision of
RFC 1404 will be produced as an Internet-Draft.  It will include the
above changes (and their corresponding changes to the examples in
Appendix B) and editorial improvements such as correction of typos and
polishing of the grammar.



The ``Statistics for Network People'' Document

This topic precipitated a discussion on the direction and purpose of the
working group.  This was agreed to be the development of good
general-purpose network traffic planning tools, so as to make it simpler
to produce one-off reports for problem solving as well as routine ones
for long-term purposes.

Everyone agreeed that it would be very useful to have a document which
set out our recommendations on exactly what reports are most useful for
various kinds of links.  This would include a list of variables and
their most useful sampling frequencies, together with comments on what
effects one should look for when monitoring performance.  This document
will set out our `Statistics for Network People.'

To launch this, everyone attending the meeting was asked to produce a
page or so of this kind of information and post it to the mailing list.
Nevil will collect this and edit it into a preliminary draft for further
discussion and expansion.



Revised OPSTAT Goals and Milestones

   o November 94 - Publish Internet-Draft of updated RFC 1404

   o November 94 - Publish Internet-Draft of operational statistics
     server

   o November 94 - Publish initial Internet-Draft of ``Statistics for
     Network People''

   o January 95 - Submit first two of above to IESG as Informational
     RFCs

   o March 95 - Revise ``Statistics for Network People'' Internet-Draft

   o March 95 - Discuss experiences with RFC 1404 file format



Fractal Behaviour of Ethernet Packets

Nevil Brownlee gave a brief presentation on a project at the University
of Auckland which is continuing work on this topic published over the
last four years.  The goal here is to better understand the behaviour of
packet network traffic and to make a mathematical model of it which is
simple enough to use for real-time performance measures in real time.