Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 29 Jan 2001 19:40:59 -0600
From:      David Kelly <dkelly@hiwaay.net>
To:        Tony Landells <ahl@austclear.com.au>
Cc:        "John Bolster" <j.bol@gte.net>, "Freebsd-Questions@Freebsd. Org" <freebsd-questions@FreeBSD.ORG>
Subject:   Re: collisions 
Message-ID:  <200101300140.f0U1exG06901@grumpy.dyndns.org>
In-Reply-To: Message from Tony Landells <ahl@austclear.com.au>  of "Tue, 30 Jan 2001 09:05:58 %2B1100." <200101292205.JAA20561@tungsten.austclear.com.au> 

next in thread | previous in thread | raw e-mail | index | archive | help
Tony Landells writes:
> As Kevin said, this is all pretty normal.
> 
> To give you some indication, most people consider anything up to
> between 5 and 10% collisions (compared to output packets) normal.
> 
> Although, as one network course instructor said--too many collisions
> is when the users start complaining about slow network performance.

To give you some kind of indication, this person considers anything 
under 150% to be normal and 200% is only an eyebrow lifter.

As to "too many collisions is when the users start complaining about
slow network performance", users will complain about any numbers they
can find. Such as a network statistic labeled "collisions"... :-)

There are two kinds of collisions, early and late. Those reported by 
"netstat -in" are "early". Some ethernet chipsets don't report early 
collisions at all. Late collisions are very bad and reflect serious 
network problems.

When your ethernet card is in half duplex mode and transmitting it is 
also listening. If it fails to copy the bits its sending then it knows 
there is a problem. If within the first 64 octets of an ethernet packet 
then this is an "early collision". Responds by backing off a random 
time and waiting for a clear net to try again. 

An early collision simply means two or more hosts decided to transmit at
the roughly the same time. The speed of light is what limits the length
of a network segment. It takes a while to get from one end to the other.
So a time of 64 octets is given for overlap. Anything beyond 64 octets
is a late collision indicating your network is too long or a host's
network stack is badly broken.

File transfer between two fast hosts on a half duplex network will
result in a collision for most every packet. The sending host will be
dumping packets on the wire tail-up with no pauses between. The
receiving host (being another fast computer) will calculate checksum on
incoming packet, queue an ACK, and try to send it on the next time
frame. But the sender is already sending the next data packet.
Collision. One for most every packet.

An SGI employee used to have a web page addressing the issue with real
mathematical analysis but I've lost the reference and not been able to
find it again. The number that comes to mind is for collision rate less
than 100% (or 150% or 200%, I forgot what threshold) the 10baseT network
thruput dropped only 8%.


--
David Kelly N4HHE, dkelly@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-questions" in the body of the message




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200101300140.f0U1exG06901>