Subject: Re: Linux seeming to run faster?
To: Matthew Orgass <darkstar@pgh.net>
From: Greg Lehey <grog@lemis.com>
List: tech-kern
Date: 10/10/2000 18:26:08
On Sunday, 8 October 2000 at 3:22:10 -0400, Matthew Orgass wrote:
> On Sat, 7 Oct 2000, Greywolf wrote:
>
>> I have a question that's been kind of burning on me for a while now,
>> and that is what in blazes are they doing that is giving the illusion
>> that they run faster than we do on just about every platform?
>
> Linux probably does run faster on every platform. Just try to read the
> source to see why: most of the OS is MD and must be rewritten for each
> port. In many places, Linux directly uses MD instructions where NetBSD
> has several function calls. In other places, Linux structures represent
> registers where NetBSD structures represent whatever the registers mean.
> Porting Linux almost means writing a new OS.
>
> In return, the time it takes to port NetBSD is a small fraction of the
> time it takes to port Linux, and once that is done one hour of MI work on
> NetBSD is usually worth at least 1 * N ports hours work on Linux and often
> much more. There are far more Linux developers than NetBSD developers,
> yet NetBSD runs better on more platforms.
I think you're missing something here: the speed of the OS is not just
in the low-level code. Sure, you can file a few instructions here or
there by writing things MD, but the real issue is in more complicated
structures like the VM system.
>> Tried Linux. Want NetBSD. But what gives in the performance
>> department?
>
> You shouldn't see a huge performance difference unless something is
> wrong. A 5% difference on modern hardware would be reasonable (with some
> time made up by better algorithms).
I think before you get too concerned, you should measure the
performance. There's been a lot of FUD spread in all directions by
people who say "my <foo> works better than your <bah>" (yours is one
of the few that turns this around and says "my <foo> works worse than
your <bah>"). I've been meaning to do some real tests some day, but
so far it's waiting on a tuit.
> However, NetBSD has been getting slower and should probably be
> audited for reasonable performance enhancements and delay abuses.
This sounds reasonable, if only to confirm whether it's true or not.
Greg
--
Finger grog@lemis.com for PGP public key
See complete headers for address and phone numbers