Article View: demon.tech.pc
Article #34225Re: Ethernet Cable Length
From: Dale Shuttleworth
Date: Sun, 18 Apr 1999 00:00
Date: Sun, 18 Apr 1999 00:00
79 lines
4403 bytes
4403 bytes
Hi,
Richard Ashton ({R}@sunshine.tm) wrote:
: In article <y$4ObJAcKIG3EwQE@ch-w.demon.co.uk>
: jc@ch-w.demon.co.uk "John Chapple" writes:
:
: [...]
:
: } Not really related to subject, but, I understand that on an ethernet
: } network if utilization goes above 25% this is regarded as 'a bad thing'
: } (whatever that means) my 3com hub has peaked at 80% whilst installing a
: } sp over the network on 2machines, just using netbeui BTW. Would this be
: } a matter of concern on a real network?
No, providing you are using a sensible packet size (MTU in TCP/IP).
: A bad thing in ethernet terms are "collisions", caused when two NICs
: start talking at the same time. Each sees the resulting corruption and
: backs off for a random time. If too many talkers are trying to talk at
: the same time the collisions go up and throughput down, dramatically.
Collisions are normal for medium to highly loaded networks with more
than two hosts, fairly inevitable really - if all hosts on a three
host network are trying to transmit, two will be ready to transmit as
soon as the other is finished and a collision will occur.
: So there is a problem as utilisation and collisions rise, if
: uncontrolled the lan will end up all collisions and no throughput.
This assumes that you have broken ethernet adaptors which do not
correctly back off or you have not followed the rules and your collision
domain is too large. Alternatively, you have set a painfully small
maximum packet size. The first two mean that your network is broken and
it is only luck that it works at all, the other means that your network
administrator needs re-educating, your software is broken, or you are
trying to do something really unusual (ATM cells over ethernet anyone?).
: On a 2 machine network there is not another talker trying to speak so
: in this rare case 100% utilisation is OK if perhaps in practice
: unachievable..
:
: On a multistation network with a number of active talkers 80% would
: not be reached because of collisions.
Not true. Collisions have a pretty marginal effect on network capacity
in most cases. The collision time is very short and providing the
packet length is comparitively long, high utilisations are easily
achievable.
: Token Ring does not suffer from colisions and can happily be run at
: 80%-90% utilisation. The network manager would be advised to address
: the issue, as it can only ever get worse.
Ethernet can also happily be run at 80-90% utilisation. Of course,
this is only practical if you are using large packet sizes. If your
overlying protocol / application uses 100 byte packets you won't do
very well. If you're using packet sizes of 500 or 1500 bytes you'll
do a lot better.
There is a lot of rubbish talked about ethernet capacity. 37%
utilisation is the *theoretical* *worst case* capacity. Most people
will do an *awful* lot better than that.
Of course, if you have a network which is normally running at 80%
capacity it probably needs upgrading, not because of the networking
technology used, but simply because in six months you are likely to
have a network running at 100% (effective) capacity and people will
start to complain about the speed.
For a lot more information try http://www.ots.utexas.edu/ethernet/
it will answer (or point you to the answer to) most ethernet
questions.
Dale.
--
**************************************************************************
* Dale Shuttleworth *
* Email: dale@giskard.demon.co.uk *
**************************************************************************
Technical Details
Message-ID:
<FAE35u.uv@giskard.demon.co.uk>Path:
arc1-ca.pugleaf.net!unknown!not-for-mailReferences:
<7ww24BAvpgD3MA5U@electrue.co.uk> <y$4ObJAcKIG3EwQE@ch-w.demon.co.uk> <924362296snz129.os2.7@sunshine.tm>Show Full Headers
Complete Article Headers
X-Google-Language: ENGLISH,ASCII-7-bit X-Google-Thread: 1036bc,5bf495d5beca8be8 X-Google-Attributes: gid1036bc,public From: dale@giskard.demon.co.uk (Dale Shuttleworth) Subject: Re: Ethernet Cable Length Date: 1999/04/18 Message-ID: <FAE35u.uv@giskard.demon.co.uk> X-Deja-AN: 468085309 X-NNTP-Posting-Host: giskard.demon.co.uk:158.152.43.164 References: <7ww24BAvpgD3MA5U@electrue.co.uk> <y$4ObJAcKIG3EwQE@ch-w.demon.co.uk> <924362296snz129.os2.7@sunshine.tm> X-Complaints-To: abuse@demon.net X-Trace: news.demon.co.uk 924476536 nnrp-12:11040 NO-IDENT giskard.demon.co.uk:158.152.43.164 Organization: Dale's home Linux box. Newsgroups: demon.tech.pc