From mboxrd@z Thu Jan 1 00:00:00 1970 From: Michael Tokarev Subject: Re: kvm guest: hrtimer: interrupt too slow Date: Sat, 10 Oct 2009 13:18:16 +0400 Message-ID: <4AD05158.8030503@msgid.tls.msk.ru> References: <20091003231205.GA15015@amt.cnet> <20091007231733.GG5903@nowhere> <20091008005456.GA10032@amt.cnet> <4ACD9AB9.3080803@msgid.tls.msk.ru> <4ACD9F68.10303@msgid.tls.msk.ru> <4ACDF1EF.3070005@msgid.tls.msk.ru> <20091008195232.GA8350@amt.cnet> <4ACFA988.6070500@msgid.tls.msk.ru> <20091009222733.GC5859@nowhere> <4ACFBA6E.7030901@msgid.tls.msk.ru> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Cc: Marcelo Tosatti , Thomas Gleixner , kvm To: Frederic Weisbecker Return-path: Received: from isrv.corpit.ru ([81.13.33.159]:36578 "EHLO isrv.corpit.ru" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754043AbZJJJSx (ORCPT ); Sat, 10 Oct 2009 05:18:53 -0400 In-Reply-To: <4ACFBA6E.7030901@msgid.tls.msk.ru> Sender: kvm-owner@vger.kernel.org List-ID: Michael Tokarev wrote: > Frederic Weisbecker wrote: [] >>>> Was there swapping going on? >>> Not as far as I can see, and sar output agrees. >> >> But I can read this from you guest traces: I missed this one yesterday. Note it's GUEST traces indeed. Higher (read: non-zero) pgp{in,out} and faults values happens in *guest*, not on host (original question was if we've swapping in HOST, which'd explain the timer issues) [cutting extra all-zero columns] >> pgpgin/s pgpgout/s fault/s pgfree/s >> 11:44:47 0.00 32.32 907.07 277.78 >> 11:44:48 27.59 22.99 44.83 150.57 >> 11:44:49 0.00 33.68 22.11 218.95 >> [...] >> 21:46:54 0.00 31.68 16.83 90.10 >> 21:46:55 0.00 108.00 17.00 89.00 >> 21:46:56 9.76 482.93 3890.24 439.02 >> 21:46:57 0.00 760.00 8627.00 1133.00 >> 21:46:58 0.00 84.85 2612.12 138.38 >> 21:46:59 0.00 16.00 17.00 90.00 >> >> So it looks like there was some swapping in when the hrtimer (spuriously) >> hanged. One possible guess. Since the guest hanged for some time, the higher values there might be a result of accumulated values for several seconds. > It's not swapping. Swapping is in a separate table, with columns titled > pswpin/s and pswpout/s -- first table. > > On my home machine with no swap at all, 4gig memory and 2gig free, > pgpgin and pgpgout are increasing too. > > Also, while in the second case (21:46:56) there's actually some > noticeable activity (page faults at least), in first case that > activity is modest. > > Note there's no documentation for /proc/vmstat file :) > > /mjt