Re: [kvm-devel] [Virtio-for-kvm] [PATCH 7/7] userspace virtio
Anthony Liguori wrote: I think we should hold off on this sort of patch at first. I know it improves performance, but it's very hack-ish. I have a similar patch[1] that improves performance more but is even more hack-ish. I think we have to approach this by not special cases virtio-net to know about the tap fd, but to figure out the interface that virtio-net would need to be efficient, and then refactor the net interface to look like that. Then we can still support user, pcap, and the other network transports. [1] http://hg.codemonkey.ws/qemu-virtio/file/75cefe566cea/aio-net.diff While you are right in principle, high performance networking is long overdue in kvm so I applied that patch. Once a mega async dma framework is added to qemu, we'll just revert that patch prior to adding the glue to said framework. -- error compiling committee.c: too many arguments to function - This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2005. http://clk.atdmt.com/MRT/go/vse012070mrt/direct/01/ ___ kvm-devel mailing list kvm-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/kvm-devel
Re: [kvm-devel] [Virtio-for-kvm] [PATCH 7/7] userspace virtio
Avi Kivity wrote: Anthony Liguori wrote: I think we should hold off on this sort of patch at first. I know it improves performance, but it's very hack-ish. I have a similar patch[1] that improves performance more but is even more hack-ish. I think we have to approach this by not special cases virtio-net to know about the tap fd, but to figure out the interface that virtio-net would need to be efficient, and then refactor the net interface to look like that. Then we can still support user, pcap, and the other network transports. [1] http://hg.codemonkey.ws/qemu-virtio/file/75cefe566cea/aio-net.diff While you are right in principle, high performance networking is long overdue in kvm so I applied that patch. Once a mega async dma framework is added to qemu, we'll just revert that patch prior to adding the glue to said framework. I second Avi, the reason we wanted quick (and ugly) performance patch merged was because people are trying kvm and sometime get disappointed of IO performance. Since we did have unmerged pv net for quite a while, we're pushing towards merging it. We can't wait for qemu guys to merge it and we also plan to develop host kernel pv network side. btw: My hack leaves the -user support but with the previous performance. I'll add the tx improvement too. Thanks for the backward compat patch :) Dor - This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2005. http://clk.atdmt.com/MRT/go/vse012070mrt/direct/01/ ___ kvm-devel mailing list kvm-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/kvm-devel
Re: [kvm-devel] [Virtio-for-kvm] [PATCH 7/7] userspace virtio
Dor Laor wrote: From f244bcad756c4f761627557bb7f315b1d8f22fb2 Mon Sep 17 00:00:00 2001 From: Dor Laor [EMAIL PROTECTED] Date: Thu, 20 Dec 2007 13:26:30 +0200 Subject: [PATCH] [VIRTIO-NET] Rx performance improvement The current performance are not good enough, the problem lies in qemu tap handling code that caused to pass packets one at a time and also to copy them to a temporal buffer. This patch prevents qemu handlers from reading the tap and instead it selects the tap descriptors for virtio devices. This eliminates copies and also batch guest notifications (interrupts). Using this patch the rx performance reaches 800Mbps. The patch does not follow qemu's api since the intention is first to have a better io in kvm and then to polish it correctly. This breaks -net user, which is one of the motivations for having a qemu net device. We need to maintain the old path for that, and only use the new fast path if using -net tap. -- error compiling committee.c: too many arguments to function - This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2005. http://clk.atdmt.com/MRT/go/vse012070mrt/direct/01/ ___ kvm-devel mailing list kvm-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/kvm-devel