From mboxrd@z Thu Jan 1 00:00:00 1970 From: Christian Theune Subject: CPU usage wrt kernel mode Date: Fri, 05 Sep 2008 09:23:36 +0200 Message-ID: <1220599416.25056.17.camel@mindy> Reply-To: ct@gocept.com Mime-Version: 1.0 Content-Type: multipart/signed; micalg=pgp-sha1; protocol="application/pgp-signature"; boundary="=-E/VQzXq8nPS1VKxRAJdD" To: kvm@vger.kernel.org Return-path: Received: from mail.gocept.net ([213.187.85.171]:49172 "EHLO mail.gocept.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751621AbYIEHdU (ORCPT ); Fri, 5 Sep 2008 03:33:20 -0400 Received: from localhost (localhost [127.0.0.1]) by mail.gocept.net (Postfix) with ESMTP id CFE84305E576 for ; Fri, 5 Sep 2008 09:23:37 +0200 (CEST) Received: from mail.gocept.net ([127.0.0.1]) by localhost (mail.gocept.net [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id 0ZWDWf4KH8ft for ; Fri, 5 Sep 2008 09:23:37 +0200 (CEST) Received: from [195.62.106.86] (unknown [195.62.106.86]) (Authenticated sender: ct@gocept.com) by mail.gocept.net (Postfix) with ESMTPSA for ; Fri, 5 Sep 2008 09:23:37 +0200 (CEST) Sender: kvm-owner@vger.kernel.org List-ID: --=-E/VQzXq8nPS1VKxRAJdD Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: quoted-printable Hi, we're setting up a couple of machines with KVM and are really happy in general. However, we have a consistently appearing issue with a high percentage of CPU time spent in kernel (aka system) mode. We're using kernel 2.6.25 on the host and 2.6.25 in the guest. KVM 73 is compiled manually on the host. We eliminated virtio as a problem source already, as well as specific disk storages (local LVM device, AoE device, qcow and raw images). We use a Linux kernel compilation run to produce the behaviour by issuing `make -j(N+1)` (for N CPUs) and comaring various virtual machine configurations as well as the host. The hosts have 4 to 8 cores and the guests were run with 1 or 2 virtual CPUs. In a guest with a single CPU the percentage of system mode time goes up to 70-80% grinding the VM almost to a halt. On the host system load stays around 10%. Latencytop seems to indicate file operations that might cause the CPU time. The filesystem is ext3 on the host and the guests. We're kind of stuck after a couple of days of debugging and wonder whether someone knows this behaviour (we assume this isn't supposed to be this way) and might know how to debug it further (or even solve it). Best regards, Christian --=20 Christian Theune =C2=B7 ct@gocept.com gocept gmbh & co. kg =C2=B7 forsterstra=C3=9Fe 29 =C2=B7 06112 halle (saale= ) =C2=B7 germany http://gocept.com =C2=B7 tel +49 345 1229889 7 =C2=B7 fax +49 345 1229889 1 Zope and Plone consulting and development --=-E/VQzXq8nPS1VKxRAJdD Content-Type: application/pgp-signature; name=signature.asc Content-Description: This is a digitally signed message part -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) iD8DBQBIwN54dUt9X/gknwIRAmYuAKDT379ubswRo5EHTD0DiIn0APuTXQCgiLmM P2U5ZhYLu3LHVgUtBxtvGpQ= =KjBB -----END PGP SIGNATURE----- --=-E/VQzXq8nPS1VKxRAJdD--