Hello everybody, i just wondered why my xen-maschine is ~37 times slower in starting processes as my workstation. workstation: time for id in $( seq 0 1000 ); do /bin/true; done real 0m1.358s user 0m1.142s sys 0m0.090s time /bin/true real 0m0.002s user 0m0.001s sys 0m0.001s xen 3.4.3 (2.6.18.8-xen, Lenny, AMD II X4, 16G RAM): time for id in $( seq 0 1000 ); do /bin/true; done real 0m49.318s user 0m2.090s sys 0m41.400s time /bin/true real 0m0.038s user 0m0.010s sys 0m0.010s so my xen-hosts needs ~37x more time to complete the loop and even a simple /bin/true is >10x slower. The native CPU-performance is not affected, compressing/openssl-benchmarks are running fine. Looks like the host have some troubles in starting/switching processes. Any ideas where to look at/start? Thanks a lot, Fabian
Hi,> Any ideas where to look at/start?please start by doing the comparisms using the same system, and then look into different domU kernels. This, for example is on my irc VM, running 2.6.26-2 debian kernel on a squeeze host. The VM is actually even cpu capped to 70% of one i7 core. $ time for id in $( seq 0 1000 ); do /bin/true; done real 0m1.14s user 0m0.21s sys 0m0.57s If it''s taking 40s(!) for you then start looking for disk trashing in the host, too. i.e. md_resync used to completely abuse servers while it was running. Greets, Florian
Hi 2011/12/25 Florian Heigl <florian.heigl@gmail.com>:> please start by doing the comparisms using the same system, and then > look into different domU kernels.Quite hard, because the affected system is running prod. services. FYI: I didn''t run the tests in DomU, I tested Dom0!> This, for example is on my irc VM, running 2.6.26-2 debian kernel on a > squeeze host. > The VM is actually even cpu capped to 70% of one i7 core.But here the tests done in DomU: DomU - Squeeze 2.6.32-5-xen-686 real 1m3.636s user 0m1.192s sys 1m3.356s DomU - Etch 2.6.18-6-xen-686 real 0m48.190s user 0m0.208s sys 0m47.271s> If it''s taking 40s(!) for you then start looking for disk trashing in > the host, too. > i.e. md_resync used to completely abuse servers while it was running.Host is powered by (real) hardware-raid, so md_resync is impossible. Also checked io-usage, but nothing strange here. Also limited prio of DomU-io-processes in Dom0 to IDLE, didn''t help. I think I have to reproduce this on different hardware to get this solved. Any further hints would be nice. Thanks a lot, Fabian
> > > I think I have to reproduce this on different hardware to get this solved. > > Any further hints would be nice. > > Thanks a lot, > > Fabian > > My prediction is on different hardware you won''t see it.Grant McWilliams http://grantmcwilliams.com/ Some people, when confronted with a problem, think "I know, I''ll use Windows." Now they have two problems. _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Hello, you are right, I already tried to update Xen to current Version 4.1.2 and used latest Squeeze-Kernel, but still the same. Maybe interessting: The problem doesn''t occour if plain Squeeze-Kernel is running (without Xen) Any ideas? Thanks a lot, Fabian 2011/12/26 Grant McWilliams <grantmasterflash@gmail.com>:>> >> I think I have to reproduce this on different hardware to get this solved. >> >> Any further hints would be nice. >> >> Thanks a lot, >> >> Fabian >> > My prediction is on different hardware you won''t see it. > > > Grant McWilliams > http://grantmcwilliams.com/ > > Some people, when confronted with a problem, think "I know, I''ll use > Windows." > Now they have two problems.
Hello, here my dmesg-logs. Maybe somebody is able to give me a hint? Thanks a lot, Fabian Zimmermann _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Hello again, Just detected: "events/0" is running all the time ~25% cpu-load. Looks like there is no "quick fix", isn''t it? Thanks, Fabian
> Looks like there is no "quick fix", isn''t it?Alright, you tested dom0. Are any domU''s online during that test?
Hello, 2011/12/31 Mark van Dijk <mark@internecto.net>:> Alright, you tested dom0. Are any domU''s online during that test?Yes, I tested Dom0 and DomUs, both are affacted. To answer your question: The loop is slow even if there is no DomU running, but I didn''t test if events/0 is using too much CPU if there are no DomUs running. So: - slow loop in Dom0/DomU regardless of DomUs are running or not. - high events/0 cpuload if DomU is running (didn''t test without running DomU). Some - maybe usefull - informaton: I also detected a very slow vmlinuz-loading/initrd-processing while booting Dom0. Thanks a lot, Fabian Zimmermann
Fabian Zimmermann <xenusers@z-technologies.de> writes:> my xen-hosts needs ~37x more time to complete the loop and even a > simple /bin/true is >10x slower. The native CPU-performance is not > affected, compressing/openssl-benchmarks are running fine.Have you got the libc6-xen package installed? -- Regards, Feri.
On Sat, Dec 31, 2011 at 09:38:50AM +0100, Fabian Zimmermann wrote:> Hello, > > 2011/12/31 Mark van Dijk <mark@internecto.net>: > > Alright, you tested dom0. Are any domU''s online during that test? > > Yes, I tested Dom0 and DomUs, both are affacted. > > To answer your question: > > The loop is slow even if there is no DomU running, but I didn''t test > if events/0 is using too much CPU if there are no DomUs running. So: > > - slow loop in Dom0/DomU regardless of DomUs are running or not. > - high events/0 cpuload if DomU is running (didn''t test without running DomU). > > Some - maybe usefull - informaton: I also detected a very slow > vmlinuz-loading/initrd-processing while booting Dom0. >From the dmesg you posted I see your system is AMD based. There''s a known bug causing big performance hit on AMD systems, and there''s a patch for it available here: http://xenbits.xen.org/hg/xen-4.1-testing.hg/rev/435493696053 You need Xen with that patch included. The patch is included in the latest Xen 4.1.3 release. -- Pasi