On Mon, Oct 09, 2006 at 10:16:05PM +0200, Sebastien Godard wrote: > Bryce Harrington wrote: > >On Sun, Oct 08, 2006 at 03:40:56PM +0200, Sebastien Godard wrote: > > > >>To know exactly how things work, could you please do the following and > >>then send me the various snapshots you got: > >>0) CPU's 0 and 1 online. > >>1) Take snapshot #1 of /proc/stat file > >>2) Disable CPU 1 > >>3) Take snapshot #2 of /proc/stat > >>4) Wait a few (2-3...) seconds > >>5) Take snapshot #3 of /proc/stat > >>6) Enable CPU 1 > >>7) Take snapshot #4 of /proc/stat > >> > > > >I've done the above for two different systems for comparison. The > >capture script is also attached. > > > Thanks a lot. This will be useful for me to know how to update sysstat code. Sure, let me know if you need any other info. > >>Also could you tell me if there is a specific processor (for example CPU > >>0) that can *never* be disabled? > >> > > > >No, but this is architecture dependent. All systems prevent you from > >offlining ALL cpus, but some (like em64t) restrict offlining cpu0, while > >others (ppc64 for example) simply require that any one cpu remain > >online, and do allow offlining cpu0. > > > This is a real difficulty... sar has to compute the interval of time > elapsed since its previous sample, and needs for this a processor that > has never been disabled during this interval of time. So this will be a > limit for sar, which will assume that one cpu (cpu0) can never be disabled. For my own purposes, that solution should work just fine. Since some architectures don't allow hotplugging cpu0 at all, it seems like a very reasonable pre-condition for sysstat. > I'm going to update sysstat in the coming days (or weeks, depending on > the free time I have). I will submit it to you so that you can try it on > your systems. Great, thanks, and I'll be happy to test it for you. :-) Bryce
