Date: Thu, 20 Oct 2005 08:34:38 -0600 From: Scott Long <scottl@samsco.org> To: John Baldwin <jhb@freebsd.org> Cc: src-committers@freebsd.org, Andrew Gallatin <gallatin@cs.duke.edu>, Bruce Evans <bde@zeta.org.au>, cvs-src@freebsd.org, cvs-all@freebsd.org, David Xu <davidxu@freebsd.org> Subject: Re: cvs commit: src/sys/amd64/amd64 cpu_switch.S machdep.c Message-ID: <4357AAFE.2070002@samsco.org> In-Reply-To: <200510200958.09182.jhb@freebsd.org> References: <200510172310.j9HNAVPL013057@repoman.freebsd.org> <4355080C.302@samsco.org> <20051020145234.H99720@delplex.bde.org> <200510200958.09182.jhb@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
John Baldwin wrote: > On Thursday 20 October 2005 01:45 am, Bruce Evans wrote: > >>On Tue, 18 Oct 2005, Scott Long wrote: >>I use 100 and never downgraded to use 1000 except for testing how bad >>it is. The default number is now up to <number of CPUs> * 2 * HZ. >>E.g., it is 4000 on sledge.freebsd.org. While 4000 interrupts/sec can >>be handled easily by any new machine, 4000 is a disgustingly large >>number to use for clock interrupts. Have a look at vmstat -i output >>on almost any machine. On most machines in the freebsd cluster, the >>total number of interrupts is dominated by clock interrupts even with >>HZ = 100. > > > Note that on 4.x you don't get to see the interrupt counts for the hz + stathz > * (cpus - 1) IPIs for all the clock interrupts, so in real numbers, each CPU > has gone from hz + stathz to hz * 2 interrupts. However, the higher number > is offset by the fact that the interrupt handler for the lapic case doesn't > have to touch any hardware, and it also works much more reliably (getting > irq0 to work in APIC mode on some amd64 nvidia chipsets required several > quirks, and future motherboards will probably continue to require quirks > since Windows uses the APIC timer in APIC mode and doesn't require irq0 to > work in APIC mode). > I'm in complete argreement that using the APIC timer is the right thing to do, and I believe that we did some tests to show that the high interrupt rate didn't have an appreciable effect on performance. However, I'd like to revisit the HZ=1000 decision for 7-CURRENT. Scott
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4357AAFE.2070002>