[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Collisions on lan using Linux versus Windows



Bob Proulx, 2002-Sep-16 17:03 -0600:
> Jeff <jcoppock1@attbi.com> [2002-09-16 14:19:34 -0700]:
> > Robert Ian Smit, 2002-Sep-16 22:25 +0200:
> > > I suppose so. Is it still true that on a busy lan you only get 40%
> > > or less troughput? The guy who told me this years ago said that
> > > ethernet was dead and tokenring the thing to have since you're
> > > throughput would always be near 100%. Aesthetically ethernet was
> > > never a thing of beauty, but it's cheap and it works.
> > 
> > On a busy Ethernet segment that's shared, not switched, it's more like
> > 60% throughput.  The collisions and subsequent back-off routines
> 
> But if you keep putting more and more talkers on the bus you will
> eventually see even more decay of performance.  Because it is a
> collision detect and backoff process it is load dependent.  Back in
> the days of coax we would meltdown to around 40% before things
> stabilized.  I agree the 60% number is probably more typical of a
> badly loaded ethertwist lan.  I have seen 65% considered normal.

Good point.  I now recall my time on a 1500 meter ThickNet segment
running 10MB/sec Ethernet.  It was horrid!  I seem to recall it ran at
about 50% with only 100 talkers.

> > Token Ring is yet another example where it doesn't matter if the
> > technology is better.  I think the "cheap and it works" is what did
> > the trick for Ethernet.
> 
> Agreed.  Also you might remember a 100VG protocol which was a
> collisionless protocol.  It would give you sustainable bandwidth
> utilization regardless of load and the number of transceivers on the
> bus.  But it could not compete with 100baseT being mostly a simple
> frequency push of 10baseT.  I think the hardware was the same price.
> It just scared people that it used a slightly different technology
> than they were used to.  Also for some reason it was only available on
> MS-Windows, where network performance was not considered critical, and
> not available for the longest time on unix servers where network
> performance was considered critical.  Sigh.
> 
> Bob

I remember 100VG!  I think HP came out with that and positioned it for
multimedia with it's low-latency characteristics.  It sure sounded
cool!

jc


--
Jeff Coppock		Systems Engineer
Diggin' Debian		Admin and User



Reply to: