[RAM] RRG & ITR lists: BGP path hunting, stability, MRAI timer, CPU and memory load etc.

[Robin Whittle <rw at firstpr.com.au>]

Since the limitations of the BGP routing system are the primary
constraint forcing us to consider heroic initiatives such as
LISP, Ivip or whatever, I think many RAM list people might be
interested in some discussions concerning:

  BGP's MRAI timer and how it was specified in RFC1771 to allow
  withdrawals to pass, how this was largely ignored by router
  vendors by about 2002, and how in RFC4271 the MRAI timer was
  prohibited from allowing withdrawals to pass: the default
  timer applies to all announcements about a prefix, including
  withdrawals.

  My examples of how this leads to patterns of behavior by small
  sections of the BGP system which become "amplified" by other
  sections.  (Think "reverberated".)  A single little network of
  four routers, with modern (non-RFC1771-compliant) MRAI
  timer functions, can turn a broken link or some other
  withdrawal event into:

    1 - An announcement of a longer best path.

    2 - About 30 seconds later, the withdrawal of that prefix.

  It is easy to create a topology which produces multiple such
  announcements at ~30 second spacings, followed by the
  withdrawal 30 seconds after the last best path announcement.

  The exact timings and behavior would be more varied than
  my hand-worked examples, but the basic pattern is clearly part
  of the "path hunting" problem, as Geoff Huston's research
  indicates.

  Tony Li's and now Geoff Huston's ID on Improving BGP
  stability, which includes a "Path Length Damping" function
  which would replace the current MRAI timer with something
  (I think) not unlike what was specified back in 1995 by
  RFC1771.  Version 01 of 13 June is at:

    http://tools.ietf.org/html/draft-li-bgp-stability

  Questions about how different trends in the topology of the
  Internet affect the load on BGP routers.

  Actual capacity of routers to handle large numbers of routes
  including splitting the load amongst multiple router
  processors.  (Robert Raszuk, IDR message 2428.)

  Flap damping and how it might have been much better.

  Processor and RAM speeds, RAM capacities etc.


In the RRG archives, the starts of the threads are:

  Geoff Huston's article on BGP stability, update statistics
  and damping

    http://psg.com/lists/rrg/2007/msg00138.html   (2007-06-19)

  BGP path hunting, MRAI timer and Path Length Damping

    http://psg.com/lists/rrg/2007/msg00146.html   (2007-06-21)


In the IDR list these threads continued.  Reading messages from:

    http://www1.ietf.org/mail-archive/web/idr/current/msg02413.html

takes in the remainder of the discussion to date.

I understand the best place to discuss these things is on the
IDR list, not on the RAM list.


  - Robin             http://www.firstpr.com.au/ip/ivip/





_______________________________________________
RAM mailing list
RAM at iab.org
https://www1.ietf.org/mailman/listinfo/ram