Hi list, when upgrading from 4.8 til 4.9 on my SMP Xeon box with HTT enabled I noticed something when running top. When running 4.8 the processes seemed to switch between all four CPUs, that is both the physical and virtual CPUs. In 4.9 top only shows processes using the physical CPUs, in my case 0 and 2. Any thoughts on this? regards, -- Morten A. Middelthon I cun pleese-a oonly oone-a persun per dey. Tudey is nut yuoor dey. Tumurroo, isn't luukeeng guud ieezeer. This .sig has been encheferized - http://flipp.net/bork/
On Mon, Apr 19, 2004 at 10:58:41AM +0200, Morten A. Middelthon wrote:> top only shows processes using the physical CPUs, in my case 0 and 2. > > Any thoughts on this? > > regards, > > -- > Morten A. Middelthonsysctl machdep.halt_logical_cpus or something like that. Jiawei -- "Without the userland, the kernel is useless." --inspired by The Tao of Programming
On Mon, Apr 19, 2004 at 05:00:49PM +0800, leafy wrote:> On Mon, Apr 19, 2004 at 10:58:41AM +0200, Morten A. Middelthon wrote: > > top only shows processes using the physical CPUs, in my case 0 and 2. > > > > Any thoughts on this? > > > > regards, > > > > -- > > Morten A. Middelthon > sysctl machdep.halt_logical_cpus or something like that. > > JiaweiI found this: machdep.hlt_logical_cpus: 1 -- Morten A. Middelthon 75% ooff zee Iert is cufered by veter, Jeeps cufer zee rest. This .sig has been encheferized - http://flipp.net/bork/
Hello, On Mon, Apr 19, 2004 at 02:35:23AM -0700, Kris Kennaway wrote:> On Mon, Apr 19, 2004 at 11:20:52AM +0200, Morten A. Middelthon wrote: > > sysctl -w machdep.hlt_logical_cpus=0 > > > > that did the trick. I added it to loader.conf as well. > > > > regards, > > Note that HTT causes worse performance in many workloads, which is why > that sysctl defaults to off.Would you please point out some cases when HTT with machdep.hlt_logical_cpus=0 causes worse performance than machdep.hlt_logical_cpus=1? I am using an HTT Xeon since recently and in my tests it showed that machdep.hlt_logical_cpus=0 is better! A simple ubench (spawning 2 processes) gives better results in 0 mode! Thank you in advance! Rumen Telbizov
> Would you please point out some cases when HTT with machdep.hlt_logical_cpus> 0 > causes worse performance than machdep.hlt_logical_cpus=1?Typically, running a single, large, CPU-bound process, especially FPU-instensive simulations.