Date: Sat, 14 Dec 2002 20:36:52 -0800 From: Terry Lambert <tlambert2@mindspring.com> To: Nathan Arun <nathan_arun@hotmail.com> Cc: arch@freebsd.org Subject: Re: Threads in FreeBSD Message-ID: <3DFC06E4.9474F4C2@mindspring.com> References: <F154W6zFoVY2TwDK0qb00016b37@hotmail.com>
index | next in thread | previous in thread | raw e-mail
Nathan Arun wrote: > I'm not a kernel programmer and wouldn't know the relative merits of > different threading architectures. Purely as an FYI, here is a white paper I > came across on the internet. This is by a Redhat developer: > http://people.redhat.com/drepper/nptl-design.pdf, who argues that 1-on-1 > implementation is the best. And some benchmarks to prove it is here: > http://www.onlamp.com/pub/a/onlamp/2002/11/07/linux_threads.html?page=2 > (though this is Linux) This paper has been commented on before. The test programs used for the benchmarks are not representative of a general purpose computer with a mixed load, and only give good numbers when the system load is low and homogenous. Please see the list archives for details. If you are really interested in this, the original KSE discussions were in usenet news groups from 1994 and again in 1996. One of the participants was the architect of the threads in Solaris. The 1:1 model is definitely suboptimal, since it increases kernel/user boundary crossing frequency significantly. The Linux argument is that their implementation of this crossing has low overhead on x86 architectures. My initial complaints about the Solaris/SVR4 model in the 1994 discussion were in the context of non-use of threads for an internal product at Novell/USG (the former USL) at the time. -- Terry To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-arch" in the body of the messagehelp
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3DFC06E4.9474F4C2>