Linux Benchmarking

[Lennart Sorensen]

On Thu, Aug 18, 2005 at 12:55:55PM -0400, Evan Leibovitch wrote:
> The "how well configured" IMO is part of the original question. How well 
> are various distros tuned for certain environments? How much work do 
> they need to be reasonably optimized?

Thea features you select and the settings you make will affect it a lot.
Tuning depends on your application needs.  If there was one optimal
tuning, you would think that would be the kernel default or application
default.  Why would only one distribution have figured out that magical
tuning value that is better than all others in all cases?

> Ie, I recall that some distros were faster than others to ship with 
> kernels pre-configured to run better with Pentiums at the expense of 486 
> operation. Now that we have 64-bit coming along, some distros are now 
> shipping versions optimized for that. Yeah, anyone could eventually 
> figure out how to do this, but for the newcomer obviously it's best to 
> find something that comes out of the box reasonably well tuned.

All distributions ship with kernels optimized for different cpus.  You
just have to install the one matching your cpu.  Some libraries have
assembly optimizations in them, and at least I have seen on debian a few
like libssl where the ld-linux loader will load the one for your cpu
automatically when a program request that library.  Multiple versions of
the library are installed at once.

> Also, aren't there also some source-based distros that probe the 
> hardware and then set compile-time options based on what's found? 
> Wouldn't they run faster (all else equal) than an installation that was 
> more one-size-fits-all, in exchange for a more-complex install process?

Waste of time.  You will never gain back the wasted compile time no
matter how good those cpu specific optimizations ever get.  And gcc has
lousy optimizations in general.  Really there are not that many
different cpu types to pick from on x86, and most of them actually being
based on the previous design will like the same general kind of
optimizations.  It really doesn't make a difference in general.

Lennart Sorensen
--
The Toronto Linux Users Group.      Meetings: http://tlug.ss.org
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://tlug.ss.org/subscribe.shtml