Date: Thu, 06 Jun 2002 18:06:19 -0700 From: Terry Lambert <tlambert2@mindspring.com> To: John Baldwin <jhb@FreeBSD.org> Cc: j mckitrick <jcm@FreeBSD-uk.eu.org>, freebsd-chat@freebsd.org, "Kurt J. Lidl" <lidl@pix.net> Subject: Re: SMP/5.0 performance on single CPU? Message-ID: <3D00070B.C04A8033@mindspring.com> References: <XFMail.20020606202223.jhb@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
John Baldwin wrote: > On 07-Jun-2002 j mckitrick wrote: > >| I don't know how much overhead the locking system that FreeBSD is > >| using will impose on a uniprocessor machine, but I would guess > >| anywhere from 10%-25% slower. I think that only 10% would be pretty > >| impressive for the very first release. Again -- this is a *guess*. > > > > Any idea if a custom kernel can completely leave out the SMP code and > > end up comparable to 4.x? > > A UP kernel will leave out some of the overhead, but SMPng fundamentally > changes the way you protect data in the kernel against interrupt handlers > as well as from other threads on other CPU's (basically we are now solving > that problem in a more general case) so you can't just turn it all off. Also, for what it's worth, SVR4.2 MP was significantly faster on UP than SVR4.0.2 UP. The number one reason for this was concurrency of operations that came about from kernel preemption and subsystem reentrancy (and specifically, the FS and networking, but to a lesser extent, other subsystems). Overall, if everything ends up going correctly, turning off the SMP stuff would actually give you *worse* UP performance. -- Terry To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-chat" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3D00070B.C04A8033>