Jesse Gross wrote:
David Erickson wrote:
Jesse Gross wrote:
David Erickson wrote:
The crux is the netflow message is matching the flow expiration.
There was nothing in the ovs-vswitchd.log file other than messages
about connecting to NOX and exiting fail-open mode, which was well
before this test was performed.
Can you try turning off segmentation offload on the VM running
Apache? Something like:
ethtool -K eth0 tso off gso off
That fixed it.. is this something that needs to be done on every VM
or something that can be fixed in OVS?
I'm going to fix the stats of segmented packets in OVS so that TSO
doesn't need to be disabled.
Fantastic.
I also noticed that one of your other flows is off by the size of a
TCP SYN packet. When you respond to a packet in do you send a packet
out plus a flow mod or is it a flow mod with a buffer ID? If you send
a separate packet out then that packet is not counted as part of any
flow.
I saw this as well, I'm sure its a packet out with or without buffer,
plus flowmod. I'm fine with being a packet off, its just being an order
of magnitude off and 10 meg off in traffic is a bit of a problem ;)
Thanks for all your help in tracing down these issues.
Thank you for the fixes and great responses!
-David
_______________________________________________
discuss mailing list
discuss@openvswitch.org
http://openvswitch.org/mailman/listinfo/discuss_openvswitch.org
