From mboxrd@z Thu Jan 1 00:00:00 1970 From: Avi Kivity Subject: Re: virtio implementation? Date: Wed, 18 Jul 2007 11:49:50 +0300 Message-ID: <469DD42E.3050406@qumranet.com> References: <469C89BD0200005A00027AE1@mcclure.wal.novell.com> <64F9B87B6B770947A9F8391472E032160CC163F7@ehost011-8.exch011.intermedia.net> <1184715761.10380.37.camel@localhost.localdomain> <64F9B87B6B770947A9F8391472E032160CC169D5@ehost011-8.exch011.intermedia.net> <1184742407.10380.108.camel@localhost.localdomain> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Cc: kvm-devel-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org To: Rusty Russell Return-path: In-Reply-To: <1184742407.10380.108.camel-bi+AKbBUZKY6gyzm1THtWbp2dZbC/Bob@public.gmane.org> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: kvm-devel-bounces-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org Errors-To: kvm-devel-bounces-5NWGOfrQmneRv+LV9MX5uipxlwaOVQ5f@public.gmane.org List-Id: kvm.vger.kernel.org Rusty Russell wrote: > On Tue, 2007-07-17 at 23:49 -0700, Dor Laor wrote: > >>> On Tue, 2007-07-17 at 06:31 -0700, Dor Laor wrote: >>> >>>> btw: Rusty - what do you think of virtio for the host? >>>> >>> You mean backend? For networking it makes a great deal of sense. For >>> block it makes far less sense (COW, weird formats, etc). >>> >> I meant networking, but there are more components we can add to the >> host: >> - Block device can be directed into blkTap thus making the >> result both flexible and zero-copy. >> > > Let's avoid the term "zero-copy" since it usually indicates that the > people in the discussion can't count. Well, for block you can do zero copy, using O_DIRECT. > Let's further assume that kvm > changes to the lguest model where guest pages are user pages. > That's a valid assumption. The rest of the discussion doesn't depend on it, does it? > AFAICT the only difference between a userspace host block driver and a > kernel host block driver is now the system call overhead, and the > difficulty of doing efficient write barriers from userspace. > > True. For mortal people setups where you have at most a few thousand ops/sec, the overhead is negligible. In-kernel host drivers make sense only for 100+ disks. These people are better off using blockdevs rather than file, so an in-kernel block device need not support file-based storage. -- Do not meddle in the internals of kernels, for they are subtle and quick to panic. ------------------------------------------------------------------------- This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/