[Prev][Next][Index][Thread]
[comp.os.linux.development.system] Re: threads and Linux
I mentioned this message during the dinner and people asked me to send
it over the list (I mis-quoted some minor numbers, but the facts
presented are still impressive, IMHO)
----------------------------------------------------------------------
From: Ingo Molnar <mingo@pc5829.hil.siemens.at>
Newsgroups: comp.os.linux.development.system
Subject: Re: threads and Linux
Date: 8 Nov 1996 20:59:53 GMT
Organization: Siemens AG Austria
Message-ID: <560709$7fs@zwei.siemens.at>
References: <nitsch.846759042@mw036> <327C50E3.448F7C96@sos.net>
<y5apw1uy92l.fsf@graphics.cs.nyu.edu> <lyzg22qxliq.fsf@mouton.inria.fr>
<32824BFB.7FE6@roch803.mc.xerox.com> <lyz20e48xqp.fsf@mouton.inria.fr>
Xavier Leroy <see_my_sig_for_address@inria.fr> wrote:
: Bruce Bigby <bigby@roch803.mc.xerox.com> writes:
: > So does this mean that Linux can assign the processors of an SMP system
: > on a thread basis, rather than a process basis?
: Of course yes. In a clone()-based thread implementation such as
: LinuxThreads, threads are just processes from the kernel's standpoint.
as an addition, hard numbers about Linux kernel level threads:
on a Neptun board + 100 MHz Pentium, kernel 2.1.7:
raw two-thread hot cache context switch speed:
1.8 usecs ( 550K context switches per second)
on the same machine QNX does 230K, NT 3.51 does 90K thread context switches.
Cheers,
-- mingo
--Multipart_Wed_Nov_13_21:56:47_1996-1--