From mboxrd@z Thu Jan 1 00:00:00 1970 From: Avi Kivity Subject: Re: Support for direct inter-VM sockets? Inter-VM shared memory? Date: Sun, 23 May 2010 14:11:29 +0300 Message-ID: <4BF90D61.40206@redhat.com> References: <4BF58B3D.2090307@ncsu.edu> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Cc: kvm@vger.kernel.org, Cam Macdonell To: Tyler Bletsch Return-path: Received: from mx1.redhat.com ([209.132.183.28]:64262 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753088Ab0EWLLi (ORCPT ); Sun, 23 May 2010 07:11:38 -0400 In-Reply-To: <4BF58B3D.2090307@ncsu.edu> Sender: kvm-owner@vger.kernel.org List-ID: On 05/20/2010 10:19 PM, Tyler Bletsch wrote: > I'm interested in moving some research prototypes from Xen to KVM, but > there are a few esoteric features I'd need to do this. > > First is an efficient mechanism for direct VM-to-VM > sockets...something that bypasses the protocol stack and minimizes > overhead. Xen has XenSocket, XenLoop, and others. I found a few > mentions of this idea dating back to 2006*, mostly saying that a few > people have done something like this as a one-off, but nothing > official has been released. I haven't found anything like this more > recently, though. Has there been any progress on this front? > Ideally, I'd want a character device or a special program that acts as > a fast pipe to a different VM. > > Second, what about inter-VM shared memory? Inter-VM shared memory (contributed by Cam, copied) is in the process of being merged. That includes a fast interguest interrupt mechanism (using irqfd), so you could easily layer a fast interguest pipe on top of that. It's possible to do a dedicated guest-to-guest channel using virtio and a dma engine, but I'm not aware of any efforts in that direction. -- error compiling committee.c: too many arguments to function