Date: Thu, 04 Apr 2002 17:55:53 -0800 From: Terry Lambert <tlambert2@mindspring.com> To: Stephen J Bevan <stephen@etunnels.com> Cc: John Regehr <regehr@cs.utah.edu>, freebsd-hackers@FreeBSD.ORG Subject: Re: Linuxthreads on Linux vs FreeBSD performance question Message-ID: <3CAD0429.1A196F48@mindspring.com> References: <Pine.LNX.4.21.0204031311270.15454-100000@famine.cs.utah.edu> <3CAC036C.71DB41BB@mindspring.com> <15532.35908.341722.136026@apathy.etunnels.com> <3CACE84F.D54C9248@mindspring.com> <15532.60584.310576.443702@apathy.etunnels.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Stephen J Bevan wrote: > This is the same way that the scheduling CPU and process group > affinity crap that Linux puts up with just falls out of the > code, as well, when you go to per CPU run queues > > Since the thread on freebsd-arch didn't appear to have anything > specifically to do with Linux 2.4.17 (it was about adding VOPs in > FreeBSD) I assumed you weren't aware that the scheduler had changed > over the last few months otherwise you wouldn't have mentioned the > issue. Again, my mistake. I should have been clearer. When it comes to multiple CPUs in the same machine, what you are really trying to achieve is group negaffinity for the same CPU, in order to maximize concurrency. The intra-CPU affinity in Linux is what I am saying is bad, in that statement. The inter-cpu negaffinity isn't that hot, either. This is seperate form the issue of the act of migration of a proceess from one CPU sheduler queue to another: the negaffinity is about the decision of *when* to migrate, while the per CPU scheduling queues is about the decision of *how* to migrate. -- Terry To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3CAD0429.1A196F48>