Date: Fri, 04 Oct 1996 21:53:36 +0800 From: Peter Wemm <peter@spinner.dialix.com> To: Rob Miracle <rwm@MPGN.COM> Cc: freebsd-smp@freebsd.org Subject: Re: SMP processor utilization question Message-ID: <199610041353.VAA12611@spinner.DIALix.COM> In-Reply-To: Your message of "Fri, 04 Oct 1996 09:14:47 -0400." <3.0b24.32.19961004091446.006a680c@central.TanSoft.COM>
next in thread | previous in thread | raw e-mail | index | archive | help
Rob Miracle wrote: > We are messing with the SMP kernel on a P6/200 Dual board. I am not sure > of the board's maker. It has an AMI bios. Any way, since we upgraded the > kernel from yesterday's (and again after todays) 8AM EDT fetch of the CVS > tree, the new kernel shows (via ps aux) that (cpuidle1) is using 99.4 % of > the CPU and > (cpuidle0) is using 0% of the CPU. Vmstat shows 0% free CPU and 100% > system calls. All process start times are 31Dec69. The load average is > 1.92 sustanined. > > How can I test that both processors are working? Is this normal behavior? > I have included the output from my mptable run. > > Thanks > Rob It's unexpected, but it's broken on my machine too. The two cpuidle procs should be about evenly allocated cpu time. I'm checking into it, it's most likely something I've broken when I created the variable number of idle procs for the available cpus. We have no easy way of indicating whether both cpus are actually running, or if once has hung. We can implement some of the hacks to help, including creating a P_ONCPU flag, as well as a last-cpu-id field in the proc structure. The top port in the ports collection could have some #ifdef P_ONCPU stuff to enable it to report the cpu id number on the display. One way of seeing it in action is to do a parallel make (such as the one in the NetBSD-current tree, somebody is merging the parallel support into ours too, it's nearly ready.) When you do a "time make -j4", check the %cpu usage. If it's over 100%, you have both cpus running usefully. ie: the processes consumed (say) 15 seconds of cpu time in only 10 seconds of wall-clock time. Cheers, -Peter
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199610041353.VAA12611>