Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 18 Aug 1998 21:47:19 -0500
From:      David Kelly <dkelly@hiwaay.net>
To:        "Vadim Belman" <voland@plab.ku.dk>
Cc:        "freebsd-stable@FreeBSD.ORG" <freebsd-stable@FreeBSD.ORG>
Subject:   Re: Intel EtherExpress Pro 10/100B network collisions 
Message-ID:  <199808190247.VAA06742@nospam.hiwaay.net>
In-Reply-To: Message from "Vadim Belman" <voland@plab.ku.dk>  of "Tue, 18 Aug 1998 12:37:02 BST." <199808181036.MAA28563@plab.ku.dk> 

next in thread | previous in thread | raw e-mail | index | archive | help
"Vadim Belman" writes:
> 
> Surprisingly, we found that Intel board produces a _huge_
> number of network collisions, almost for every output
> packet.
> 
> The strange thing is that we anyhow had a terrific
> transfer rate - absolutely the same as with 3COMs
> (~920 Kb/s on a 10 mbps connection).
> 
> It is definitely a problem with the card, for we have also
> tried 3COM on the same machine, with predictable zero number
> of collisions.  There is quite likely that this is not a problem
> with a particular Intel card, for we have also tried another one 
> --- with the same bad results.

I bet the vx0 driver lies about collisions. I forget what 10/100 3Com 
PCI card I have in my FreeBSD box at work but it *never* reports 
collisions on the 10baseT network its connected to. OTOH the collision 
LED lights solid on my little office hub when the SGI O2 and FreeBSD 
box start pushing files back and forth.

Bob Metcalf, inventor of ethernet, has stated publicly something to the 
effect that his greatest regret about ethernet was naming a particular 
event, "collision".

The truth is until your collision rate gets above 100% into the 200% 
range, you don't have a problem. There was a very nice article in SGI's 
Pipeline in recent months covering ethernet "collisions" and 
documenting the math behind a collision's effect on thruput. I can't 
find the article online tonight. Must be holding my tongue wrong or 
something.

Seems like 100% collision rate results in an 8% hit on network thruput.
This is because a "collision" occurs early, something like within the
first 80 bytes of a packet frame. So the collision doesn't consume much
bandwidth. 

The Thing To Worry About are "late collisions". Late collisions are due
to broken network stacks, hardware, or networks which have been extended
beyond the lengths permitted in the ethernet spec.


--
David Kelly N4HHE, dkelly@nospam.hiwaay.net
=====================================================================
The human mind ordinarily operates at only ten percent of its
capacity -- the rest is overhead for the operating system.



To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-stable" in the body of the message



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199808190247.VAA06742>