[4035] in linux-net channel archive
Re: Question on performance
daemon@ATHENA.MIT.EDU (Dave Platt)
Fri Aug 16 03:25:48 1996
Date: Thu, 15 Aug 1996 17:27:47 -0700
From: Dave Platt <dplatt@iq.tvsoft.com>
To: chris@topdog.pas1.logicon.com (Chris Albertson)
cc: linux-net@vger.rutgers.edu
In-Reply-To: <96Aug14.193319-0400_edt.106055-17755+26@vger.rutgers.edu>
> This is a good question. "where is the overhead?"
It depends. >>grin<<.
> These systems are on a little used 10baseT network also. I never see
> speed less then 800KB/sec form Sun to Sun but from MS Windows to Sun
> it is much slower the 500 more like 200. I get about 500 between
> two high-end Linux PCs also
> Our Sun technical rep said that Suns can push data close to the ethernet
> "wire speed" but PCs only get about 1/2 this rate. This advice was in the
> context of planning a network of Sun servers and mixed Sun/PC clients.
>
> So what is Linux doing?
I've gotten transfer speeds of upwards of 900 kbytes/second going from
a Sparc to a 486-DX2/66 Linux machine, on an otherwise-idle 10BaseT
segment.
Part of the key to getting this sort of good performance is having
high-performance I/O adapters - specifically, adapters which can do
the data transfer via a fast (PCI or VLB) busmastering. Many of the
less-expensive Ethernet and mass-storage interfaces use PIO, which
requires a pretty substantial amount of the 486 or Pentium's
processing time during each data block transfer (and often requires
quite a few trips through the interrupt handler, too).
A PCI-based Linux system equipped with a busmaster PCI SCSI controller
(e.g. NCR53c810) and a busmaster Ethernet controller (e.g. PCNet-PCI
or Tulip) will probably _not_ find its network or SCSI interfaces to
be the bottleneck - the bottlenecks will be in the filesystem / block
cache and in the network stack, I think.
A system with the same motherboard and processor, but with a "dumb"
PIO Ethernet controller and a simple IDE controller (both on the ISA
bus) will probably have substantially lower throughput.
