Re: [ILUG] Comparing OS network performance

From: Niall O Broin (niall at domain linux.ie)
Date: Sat 16 Jun 2001 - 21:26:41 IST


On Sat, Jun 16, 2001 at 04:55:11PM +0100, Jeremy Smyth wrote:

> This was discussed on Slashdot as well. Good article, not very general

I didn't see the Slashdot discussions but the assertion that

> it relies on hacked versions of the kernel

is completely wrong. The article says

] The operating systems were the latest version available from a commercial
] distribution and were not recompiled (i.e., everything was tested right out
] of the box).

although it does go on to say that

] On the UNIX-style operating systems, it was necessary to tweak the kernel
] slightly to allow the use of so many connections in one process.

This "kernel tweaking" consisted of

On Linux: echo 65536 > /proc/sys/fs/file-max changes the number of
system-wide file descriptors.

On FreeBSD: Append to /etc/sysctl (or you can use sysctl -w to add these):

kern.maxfiles=65536
kern.maxfilesperproc=32768

On Solaris: Add the following to /etc/system and reboot:

set rlim_fd_max=0x8000
set rlim_fd_cur=0x8000

I'd hardly consider that advanced wizardry - it's simply tuning of kernel
variables to suit the environment in which the kernels were running.

> W2K is not very hackable...

If the ability to do above "kernel tweaking" counts as "very hackable" then
W2K is indeed very hackable - haven't you heard of the registry ? In the
infamous Mindcraft benchmarks, Mindcraft had Microsoft registry experts on
hand to wring the last ounce of performance out of Windows (was that NT or
2K - I don't remember, nor do I much care)

> The comments go on to condemn other optimisations as biassed or ignorant.

I presume you're talking about Slashdot comments. Again, I didn't see them.
But as a general rule, I'd say all's fair in OS optimisation for the
purposes of benchmarks, provided that those optimisations are available to
all and are documented.

> Still, interesting to see a benchmark that Microsoft have not paid for :-)

Indeed. It'd be even more interesting to see a rehash of the Mindcraft
benchmarks with Linux using the 2.4 kernel, the Tux web server, and the
latest Samba.

Niall



This archive was generated by hypermail 2.1.6 : Thu 06 Feb 2003 - 13:10:41 GMT