From mboxrd@z Thu Jan 1 00:00:00 1970 From: Avi Kivity Subject: Re: Network Overruns Date: Sat, 13 Sep 2008 17:44:19 +0300 Message-ID: <48CBD1C3.4020006@qumranet.com> References: <48CAA711.8060302@theplayboymansion.net> <48CB4B24.8090505@qumranet.com> <519a8b110809130118sd392d38lb11fe2fa004638c9@mail.gmail.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Cc: kvm@vger.kernel.org To: xming Return-path: Received: from il.qumranet.com ([212.179.150.194]:45098 "EHLO il.qumranet.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750936AbYIMOqz (ORCPT ); Sat, 13 Sep 2008 10:46:55 -0400 In-Reply-To: <519a8b110809130118sd392d38lb11fe2fa004638c9@mail.gmail.com> Sender: kvm-owner@vger.kernel.org List-ID: xming wrote: >> Most likely the guest's rx queue length is greater than the host's. You >> might try >> >> ifconfig vnet0 txqueuelen 1500 >> >> (and so on for every interface) >> >> >> or perhaps reducing the guests' txqueuelen. >> > > I have the same issue (overruns) and I have stalled network (in my other > report I can only unstall the network by putting the NIC in promisc back > and forth). > > The txqueuelen is indeed greater in the guest then the tap on the host, > and I also notice that it's almost ways the output NIc which get stuck. > > So changing the tap (on the host) to 1500, I can't not reproduce the hang yet > and a ping -f -s 64000 (between guests) does not produce any packet loss. > > 27810 packets transmitted, 27809 received, 0% packet loss, time 259080ms > rtt min/avg/max/mdev = 5.013/7.639/33.999/1.536 ms, pipe 3, ipg/ewma > 9.316/7.284 ms > > I hope this is the solution for my network stalls and packet loss in games. > It's not a fix, rather a workaround. Hopefully some networking guru will come up with a real fix. What NIC model are you using in the guest? -- Do not meddle in the internals of kernels, for they are subtle and quick to panic.