SUMMARY: What does the netstat(8c) collision number mean?

From: Oliver Schoett (schoett@informatik.tu-muenchen.de)
Date: Mon Dec 02 1991 - 12:50:31 CST


Last week, I asked what the collision number produced by, e. g.,
"netstat -I le0 5" means. The respondents agreed that it counts
"the number of collisions that occurred when the interface tried to
transmit a packet", and that the collisions are just the collisions
for that machine's output packets and not for the entire net.

However, one respondent says:

   "Some systems increment [the collision count] for each collision
   that has occurred (and up to 16 per packet can occur before it is
   dropped) and some increment for each packet dropped due to
   excessive collisions."

The respondents agreed that the rate collisions/output packets should
be on the order of about 1 percent, but not more than 3 percent:

   "collisions range from small fractions of a percent for lightly
   loaded nets, up to 1%, 2% or at worst 3% for heavy traffic."

   "anything more than 3% [...] is ridiculous"

   "Our worst nightmares are about networks with 10% collision rate"

   "A typical collision rate should be about 1%"

Since in our net the typical collision rate as observed by netstat is
25--50%, the respondents suggested some possible sources of the
problem.

   "Your LAN is too busy (>50% constant usage)"

   "If that's a coaxial network segment I'd check to see if it was
   overloaded, then make sure the cables are the correct impedance,
   and finally get out the TDR and start looking for wire problems."

   "One of the machines on the net may have a bad interface, or be of
   a design that violates the ethernet spec. The IBM RS6000 with a Rev
   3 ethernet board is known to cause MAJOR headaches because it
   violates the spec."

   "Your LAN chip is getting confused by the SQE signal from your
   transceiver. On SUN's you must turn off SQE (Heartbeat) otherwise
   this happens and you lose lots of thruput."

   "If you've got any repeaters in your network, make sure that the
   transceivers they are connected to have SQE disabled. That's a
   common cause of excessive collisions and network failure."

Our networking guys say that Sun's Ethernet interface does not work
correctly and produces excessive voltages that cause other controllers
to report a collision. However, our transceivers currently have SQE
turned on, and this might well have something to do with it.

Thanks to those who responded:

   spurgeon@sirius.cc.utexas.edu (Charles Spurgeon)
   Daniel Quinlan <danq@csn.org>
   sater@cs.vu.nl
   smb@ulysses.att.com
   poffen@sj.ate.slb.com (Russ Poffenberger)
   barnett@alydar.crd.ge.com (Bruce Barnett)
   rpa@dsbc.icl.co.uk (Richard P Almeida)
   hedrick@cs.rutgers.edu

Oliver Schoett Inst. f. Informatik (H2), Technische Univ. M"unchen
                   Postfach 20 24 20, 8000 M"unchen 2, Germany
schoett@informatik.tu-muenchen.de phone +49 89 2105-2390 fax -8207



This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:06:16 CDT