Bugs or non-trivial staff found:
1) new ovs do have new policy for MTU handling. Native ovs from XCP 1.1
when found ethX interface unable to change MTU to jumbo sizes, fall back
to original MTU for bridge itself. New OVS now spamming in dmesg (or
messages?) about this, but did not change.
Not a problem here, jumbo size MTU is enabled for ixgbe devices.
Vanilla ixgbe have problems with SR-IOV (enabled by default in XCP) and
jubmbo (we met that problem with OVS because ixgbe rejects to change MTU
while SR-IOV active).
Why do you use vanilla ixgbe?
Because we've getting (I'm not very familiar with tenses - not sure how
to say in English: in past we have that trouble occasionally), we've
getting periodic network connectivity loss in XCP 1.1. It was really
funny, because we using netconsole to log hosts dmesg to log server, and
I saw message from iscsi about 'no connectivity' sent by open-iscsi via
network with is 'offline'.
After upgrading to vanilla ixgbe connectivity loss completely curved,
but few more problems between XCP, OVS and ixgbe arrived (SR-IOV/jumbo
frames conflict; OVS/ixgbe problem with MTU on bridge/interface;
mandatory to disable LRO/GRO do boot).
PS We still on XCP's ixgb for client's internet bridge.
_______________________________________________
Xen-api mailing list
[email protected]
http://lists.xen.org/cgi-bin/mailman/listinfo/xen-api
