[ovs-discuss] Efficiency of OVN packet forwarding

Justin Pettit jpettit at ovn.org
Thu Aug 4 03:55:07 UTC 2016


> On Aug 3, 2016, at 6:48 PM, Wilence Yao <wilence.yao at gmail.com> wrote:
> 
> Hi, 
> I'm studying OVN these days. I found that ovn-controller issue too match flow table to br-int in compute node. Number of flows it too large。
> 
> The question is that those flow tables may reduce efficiency of flow match in packet forwarding.
> 
> I have heard that efficiency   reduce  heavily when number of flows more than 20000 in kernal.
> Has any one  make stress testing of packet forwarding in ovn environment, especially several vms on one hypervisor.

You can't really extrapolate how many kernel flows there will be based on the number of OpenFlow flows that OVN pushes to OVS.  The kernel flows are a cache of compressed flow entries based on the traffic that's actively being seen as well as the OpenFlow flows that have been configured.  I would expect there to be orders of magnitude fewer kernel flows than OpenFlow flows.  We wrote an academic paper on how we generate those flow cache entries if you're interested:

	http://openvswitch.org/support/papers/nsdi2015.pdf

If you haven't done so, I'd recommend setting up OVN and looking for yourself.

--Justin





More information about the discuss mailing list