Re: Single subnet

Vernon Schryver <vjs%rhyolite.wpd@sgi.com> Mon, 21 May 1990 06:02 UTC

Received: from merit.edu by NRI.NRI.Reston.VA.US id aa20210; 21 May 90 2:02 EDT
Received: Mon, 21 May 90 01:01:13 EST from ucbvax.Berkeley.EDU by merit.edu (5.59/1.6)
Received: from SGI.COM by ucbvax.Berkeley.EDU (5.63/1.41) id AA13701; Sun, 20 May 90 23:00:38 -0700
Received: from whizzer.wpd.sgi.com by sgi.sgi.com (5.52/900423.SGI) for @ucbvax.berkeley.edu:katz@merit.edu id AA09610; Sun, 20 May 90 23:00:53 PDT
Received: from rhyolite.wpd.sgi.com by whizzer.wpd.sgi.com (5.52/900423.SGI) for sgi.sgi.com!merit.edu!fddi id AA00849; Sun, 20 May 90 23:00:51 PDT
Received: by rhyolite.wpd.sgi.com (5.52/900423.SGI) for @whizzer.wpd.sgi.com:katz@merit.edu id AA10462; Sun, 20 May 90 23:00:48 PDT
Date: Sun, 20 May 1990 23:00:48 -0700
From: Vernon Schryver <vjs%rhyolite.wpd@sgi.com>
Message-Id: <9005210600.AA10462@rhyolite.wpd.sgi.com>
To: Dave Katz <katz@merit.edu>
Subject: Re: Single subnet
Cc: fddi@merit.edu
Status: O

> I think that the suggested ES hello timer sent by ISs should be cranked
> way down (from 600 seconds to 20 or something..).  Given the available
> bandwidth, messages that often wouldn't bother anybody.


Hummph!  
I supposed another 3,000 frames per minute in addition to the 2,000 NIF's
and who knows how many SRF's every station has to handle won't make much
difference.  5,000 f/m is only about 83 f/s.  (500 dual-MAC stations in a
wrapped ring)  Does your UNIX box notice 80 ether packets/sec?

Did you hear that X3T9.5 at Austin made the NIF repeatition rate variable,
with the express purpose of satisfying those who wanted an optional 5
second interval?  We can now have 12,000 NIF frames/min or 200 frames/sec
for SMT information, so that everyone's ring map can be no more that
5 seconds out of date.  That is in case you have a user who moves fiddles
with cables a lot, and you need to catch him in the act.

It might be prudent to be conservative with such timers.  The NIF rate is
supposed to be met only on an unloaded ring; you are not supposed to queue
NIFs if you can't send them.  The worst case delay between consecutive
opportunities to transmit is mind boggling.  The obvious calculation gives
something like 3 minutes, so huge that I'm sure I misunderstand.  It would
be unfortunate if your routing started flapping because the ring is
saturated, forcing even more traffic to handle the flapping.

vjs