I have done more or less the same thing but I switched, for various reasons, to Debian Sid with kernel 3.1.0 for my Xen hosts nodes. The issue is with the guests not the hosts. So simply upgrading your host nodes to a later kernel will not help. In our case nothing will help as we are forced to run RHEL5 guests as our production SOE. The best solution I have for RHEL guests is:
#!/bin/bash mkdir -p /etc/udev/rules.d echo "SUBSYSTEM==\"net\", ACTION==\"add\", KERNEL==\"eth*\", RUN+=\"/sbin/ethtool -K %k tx off\"" > /etc/udev/rules.d/70-ethtool.rules This is cleaner than rc.local and the ethtool support in the network up/down scripts is completely broken on RHEL (it only supports setting duplex and speed). Dave On 09/01/2012, at 7:01 AM, Hans van Kranenburg wrote: > On 1/8/12 10:33 AM, Dave Whitla wrote: >> >> The issue persists in all kernels on the 2.6 branch. I've not >> experienced it with 3.x kernels but my scripts have disabled TX >> checksumming since I first came across the issue. >> >> The problem is that many people run VMs to support "legacy" systems >> on new high-capacity hardware or in production environments where we >> are compelled to run say RHEL which doesn't ship with a kernel newer >> than 2.6.35. So while this isn't an issue with OVS it probably should >> be prominently mentioned somewhere in your doco if only to avoid it >> appearing in list posts over and over. >> >> A fix belongs in Xen's netfront driver I think. I have assumed Hans >> is using Xen. > > That's correct. > > I'm building a test setup for replacing numerous vlan-interfaces and bridges > on a Xen setup using ovs on Debian GNU/Linux stable, which has linux 2.6.32 > and Xen 4.0. > > Hans _______________________________________________ discuss mailing list [email protected] http://openvswitch.org/mailman/listinfo/discuss
