[openbeos] Re: Multiprocessing,Mutithreading - Linux vs Openbeos

  • From: François Revol <revol@xxxxxxx>
  • To: openbeos@xxxxxxxxxxxxx
  • Date: Thu, 13 Nov 2003 14:10:26 +0100

Actually, and I was quite shocked when I saw got the proof, the BeOS 
kernel is quite awful about switching time.
I recently ported the benchmarks from this article:
http://www-106.ibm.com/developerworks/linux/library/l-rt9/?t=egrL24,p=ContextSwitching
And the results aren't very good for us :)
Though the first one doesn't matter much, as it's known fact that pipes 
in BeOS are slow.
Though, I've been told the kernel never really had been optimized much, so...
Besides for such an old piece of code, it still gets it not that bad.
As I don't have Linux installed on my dual, if anyone wants to make 
comprehensive tests, feel tree to ask for the sources.

Btw, I already have something called g77 which seems to run at least here:
GNU Fortran (GCC 3.2) 3.2 20020814 (release)
Copyright (C) 2002 Free Software Foundation, Inc.
So I guess it's possible :)

François.

Selon Michael Phipps <mphipps1@xxxxxxxxxxxxxxxx>:

> You are quite correct that certain server scenarios are focused on high 
> performance. The ones you list are perfect examples. They are also, I believe
> 
> (and I am speaking somewhat outside of what I know), examples to prove my
> point. 
> These are applications where you want throughput. You have a machine running
> one 
> (or very few) applications. Task switching is very low. The machine's whole
> job 
> is to process data in and out very quickly. New jobs don't consist of
> spawning 
> new processes - they consist of using existing processes.
> 
> BeOS and OBOS, on the other hand, are tailored for very fast switching times,
> low 
> interrupt latency and highest user responsiveness. Yes, batch systems are
> very 
> fast, but that applies to daemons and already loaded applications. Their
> start 
> time and switch time tends to be a little bit longer. Often their quantum is
> set 
> higher than that of a BeOS type system.
> 
> As far as porting Fortran, it sounds like a fun job. :) It shouldn't be all
> that 
> tough - especially if there is a Fortran front end to gcc.
> 
> On 2003-11-13 at 04:20:22 [-0500], kevin.lawton@xxxxxx wrote:
> > Sorry, Michael, but I find myself disagreeing with you somewhat. I feel you
> 
> > might be confusing, or at least blending, the roles for which server and 
> > mainframes are respectively designed for. Okay, this is only my opinion,
> but I 
> > feel that the design and function of many server 'scenarios' is to provide
> 
> > quick responsiveness: domain name server, web server, file server, etc -
> their 
> > role is to provide a quick response to requests from network clients. 
> > Mainframes, on the other hand, have mostly been used for batch processing:
> a 
> > mainframe op system tends to 'get out of the way' and allow the batch
> process 
> > maximum resources in order to achieve optimum throughput. At least, that is
> how 
> > I perceive my experience.
> > By the way, I'd be interested to know if there would be any interest in a 
> > Fortran compiler to run under OpenBeOS, as this is a project I quite fancy
> 
> > doing (but only if it would be useful). As you might guess, I have used
> Fortran 
> > to a significant degree in the past and feel it still has a place in the 
> > scientific and mathematical arena.
> > Kevin.
> 
> 


-- 



Other related posts: