[ovs-dev] [PATCH 2/2] ofproto: Avoid buffer copy in OFPT_PACKET_IN path.
jt at hpl.hp.com
Tue Apr 27 17:05:38 UTC 2010
On Tue, Apr 27, 2010 at 09:42:18AM -0700, Ben Pfaff wrote:
> Jean: I finally pushed this optimization. (It's too bad it took so
That's great, thanks a lot ! I'm not worried about the time it
takes, I'm just interested in the outcome and making sure we can
continue to work together, rather than separatly.
I've done more extensive testing of the latest patch I sent
you. My test is basically running the controller in no-flow mode and
running a netperf (all packets are forwarded via the controller).
In a normal Open vSwitch deployment, with a fast CPU (Core2) for
Open vSwitch and a similar CPU for the controller, you will not notice
the difference. On the other hand, on the ProCurve switch and its tiny
CPU I see a very reliable 5% increase in performance.
The adapative expiry patch give me a further 1% increase in
this test. You got an early (broken) version of this patch somewhere
in your mailbox.
On to the latest news on my side...
I've got a slightly updated version of my QoS patch for Open
vSwitch, ask for it when you are ready to work on QoS. I've submitted
the rate-limiter part on the mailing list as a official proposal for
1.1. I'll try to push harder on the proposal.
I also have a patch for the test suite and need to find time
to send it to Greg.
Stanford is starting to test my 1.0 firmware based on OVS. We
have already identified an issue between NoX and my
firmware. I suspect nobody bothered to check NoX against Open vSwitch,
because I believe the bug would also apply to OVS.
More information about the dev