BOUNCE rtl@rtlinux.org: Approval required: Non-member submissionfrom [David Olofson <david.olofson@reologica.se>] (fwd)

[Der Herr Hofrat <der.herr@hofr.at>]

>From owner-rtl Mon Jun 25 07:56:06 2001
Received: from mail.space2u.com (sendmail@mail.space2u.com [62.20.1.135])
	by hq.fsmlabs.com (8.11.2/8.11.2) with ESMTP id f5PDu5e04991
	for <rtl@fsmlabs.com>; Mon, 25 Jun 2001 07:56:05 -0600
Received: from cutangle.admeo.se ([212.162.175.15])
	by mail.space2u.com (8.11.3/8.11.3) with SMTP id f5PDuMK16336
	for <rtl@fsmlabs.com>; Mon, 25 Jun 2001 15:56:23 +0200
From: David Olofson <david.olofson@reologica.se>
Reply-To: david.olofson@reologica.se
Organization: Reologica Instruments AB
To: rtl@fsmlabs.com
Subject: Re: [rtl]
Date: Mon, 25 Jun 2001 15:50:42 +0200
X-Mailer: KMail [version 1.1.99]
Content-Type: text/plain;
  charset="iso-8859-1"
References: <00ec01c0fb64$727d75a0$0bd2a8c0@hclt.com>
In-Reply-To: <00ec01c0fb64$727d75a0$0bd2a8c0@hclt.com>
MIME-Version: 1.0
Message-Id: <01062515504201.21588@cutangle.admeo.se>
Content-Transfer-Encoding: 8bit

On Friday 22 June 2001 23:43, chandu Reddy wrote:
> HI
>  all
>   Is there any limit on no of threads supported by a rt linux kernel
>   i mean  if there is an  big application and how many no of threads we
> can implement with out any performance over head.

There's always some overhead. (Unless you're working with a "CPU" with the 
threads built into the core - haven't seen such a design since the early days 
of digital wavetable synths, BTW...) The question is whether or not it's 
acceptable WRT available CPU power and price/performance requirements.

Anyway; Without knowing *exactly* how the RTL scheduler works, I'm still 
pretty sure that itn doesn't matter how many threads you have (as long as 
their code and data fits in memory). What matters is how often new threads 
are "woken up" (made ready for execution), and how many threads there are in 
the "runnable queue" at the time of each evaluation. (Some schedulers do the 
actual evaluation when a thread is woken up, rather than every time a
reschedule is requested - not sure about RTL.)


As to real figures; way back there was some figure like "7 kHz periodic 
thread on a Pentium CPU", which would correspond to a real work/overhead 
ratio that would be acceptable in the general case.

Having some fresh figures, and knowing what kind of ratio we're talking about 
here, would make it possible to get an idea of how many context switches per 
second one can handle at any given real work/overhead ratio.


//David Olofson --- Programmer, Reologica Instruments AB

.- M A I A -------------------------------------------------.
|      Multimedia Application Integration Architecture      |
| A Free/Open Source Plugin API for Professional Multimedia |
`----------------------> http://www.linuxaudiodev.com/maia -'
.- David Olofson -------------------------------------------.
| Audio Hacker - Open Source Advocate - Singer - Songwriter |
`--------------------------------------> david@linuxdj.com -'

----- End of forwarded message from owner-rtl@fsmlabs.com -----