SLUG Mailing List Archives
Re: [SLUG] Synchronous network ?
- To: johna@xxxxxxxxxxxxxxxxxx
- Subject: Re: [SLUG] Synchronous network ?
- From: DaZZa <dazza@xxxxxxxxxx>
- Date: Sun Sep 24 20:44:36 2000
- Cc: slug@xxxxxxxxxxx
On Sat, 23 Sep 2000 johna@xxxxxxxxxxxxxxxxxx wrote:
> Present ethernets, from what I understand, transmit on demand and
> perform a random timeout in the event of collision. This means
> throughput drops at around 60-70% utilisation (can't remember the
> exact figure).
60% is close enough. Any csma-cd based network running higher utilisation
than that is in deep kimchee.
> How about having the nodes in a cycle, where each one transmits, after
> which the next one either transmits data or a "I'm here but no data
> to transmit", so the next one could.
> This represents an overhead when there is no data, but for network
> intensive applications present over the whole network, it would mean
> that 95% - 100% network capacity could be used (assuming maxmimum
> 5% overhead), which is a lot more than, say 75% and it would mean
> the network performance would decline linearly at saturation rather
> than failing dramatically.
What you need for this is either.
a CSMA-CA based network
A switched network
In this case, the CA stands for collision avoidance - as opposed to the CD
used in ethernet, which stands for collision detection.
> Are there drivers to perform this ? What would be involved in writing
> such drivers ?
Run token ring. Or some form of token-passing network - FDDI, token ring,
a couple of others I can't remember off the top of my head.
As for doing it on "standard" ethernet - buy a switch, and reduce your
collision domain to a lower level - or forget it.