[ovs-discuss] Unexpected flow entry expiration
ajiro at mxw.nes.nec.co.jp
Mon Feb 27 21:25:34 PST 2012
I'm evaluating OVS 1.2.2.
I'd like to report an unexpected behavior of OVS.
* What you did that make the problem appear?
1. I made following bridge.
2. I added 7,000 flow entries to switch by ovs-ofctl.
Each entries have different MAC source addresses.
Entry's idle_timeout was set in 60 seconds.
hard_timeout is 0.
3. I sent frames to bridge continuously. Frames' MAC source
address was incremented in every sending for matching
to each entries.
4. Although n_packets of entries were incremented,
after 60 seconds, all flow entries were expired.
I attached log of ovs-ofctl dump-flows (showflow.log).
* What you expected to happen?
Flow entries should not be expired.
* What actually happened.
All flow entries are expired.
* The Open vSwitch version number (as output by "ovs-vswitchd
# /usr/local/sbin/ovs-vswitchd --version
ovs-vswitchd (Open vSwitch) 1.2.2
Compiled Jan 20 2012 22:56:20
OpenFlow versions 0x1:0x1
* The Git commit number (as output by "git rev-parse HEAD"),
if you built from a Git snapshot.
Not from git.
* Any local patches or changes you have applied (if any).
No patches are applied.
* The kernel version on which Open vSwitch is running (from
/proc/version) and the distribution and version number of
your OS (e.g. "Centos 5.0").
Distribution is SLES11 SP1.
Linux version 126.96.36.199-0.7-default (geeko at buildhost) (gcc version 4.3.4 [gcc-4_3-branch revision 152973] (SUSE Linux) ) #1 SMP 2010-05-20 11:14:20 +0200
* The contents of the vswitchd configuration database (usually
I attached conf.db.
* The output of "ovs-dpctl show".
I attached ovs-dpctl-show.log.
* If you have Open vSwitch configured to connect to an
OpenFlow controller, the output of "ovs-ofctl show <bridge>"
for each <bridge> configured in the vswitchd configuration
No OpenFlow controller was set.
* A fix or workaround, if you have one.
I was not able to find any workarounds.
* Any other information that you think might be relevant.
I'm guessing that flow entry statistics are not updated
at update_stats() @ ofproto-dpif.c when many statistics
are updated at same time.
I looked return value of odp_flow_cmd_fill_info() @ datapath.c
when running this test. It was failed in EMSGSIZE error.
I'm thinking that large statistics information should
be split into suitable size of netlink messages.
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 2963 bytes
Desc: not available
More information about the discuss