Re: [nsp] a question on LATE COLLISION

From: Spencer Garrett (srg@2alpha.net)
Date: Fri Mar 16 2001 - 02:26:07 EST


On Thu, 15 Mar 2001, Tatsuya Kawasaki wrote:

> C is a quite bit far way from others.
> close to 280ft via Cat 3(house pairs)
>
> others are relatively close to hub.
> All needs to talk to G to get out to internet.
>
> G log shows that
> !
> %LANCE-5-LATECOLL: Unit 0, late collision error
> !
>
> when we removed C out of net, it "appeared" to run without
> LATECOLL Errors.
> (I have moved to different place in the HUB but no changes.)
>
> If it is ineed C source of Late collision then can I duplicate
> via sending "large" packets to LAN while C is speaking.
> ie continous ping to C while sending "large" packet (extended ping)
> to any hosts in LAN.

It's *small* packets that trigger late collisions. That's the entire
reason that there is a *minimum* packet length for Ethernet. The
sender at one end of the wire can finish sending a short packet
before the collision propagates from the other end. The reason
for the minimum packet length, maximum wire length, and maximum
number of repeaters is the need to have everyone agree that a
packet has either collided or not. You have too much delay.
One possible solution would be to put a switch (not a hub)
somewhere in that 280 ft. run (preferably near the center).

Spencer



This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:32 EDT