I'm testing OVS 1.04 on a Virtualbox VM running Ubuntu 12.04 LTS with 1024 MB of RAM and I have to say that after 780 GRE port additions OVS shows no signs of slow down. We were initially trying to create a kind of GRE tunnel convergence between hosts when the network is first brought online where every host would have a tunnel interface to every other OVS host but that idea doesn't seem to scale well. The problem is due to the Bridge table becoming massive and parsing a MONITOR response, extracting UUID info etc gets increasingly slower. Not an OVS issue but interesting to me none the less. So we will have to go the route of dynamic tunnel creation and deletion when tunneling is not in use but OVS makes one heck of an encapsulator when paired with JSON RPC and OpenFlow.
On Thu, Aug 15, 2013 at 1:14 AM, Ben Pfaff <[email protected]> wrote: > On Wed, Aug 14, 2013 at 11:28:05PM -0400, Dmita Levy wrote: > > Our research meeting today ended on the question of whether or not it > would > > be possible or system breaking to tunnel to a large number of hosts using > > OVS. So I threw my JSON RPC tunnel creator into a loop and around 234 on > > bridge0 decided to stop. That being said I noticed with the ovs-vsctl > show > > command I did not see 234 gre ports (at a glance). My question is, how > many > > GRE ports can an OVS bridge have and what kind of system resources does a > > GRE tunnel port use? > > In OVS 1.10 and later, a tunnel port has pretty minimal requirements. > At a guess, a few thousand should not be a problem. > > Tunnel monitoring (CFM or BFD) is more expensive than tunnels > themselves. > > If you think there's a bug causing ports not to be created, please > report it. But ovs-vsctl just reports what's in the OVSDB database, and > OVSDB is pretty well tested, so we would need some real details. >
_______________________________________________ discuss mailing list [email protected] http://openvswitch.org/mailman/listinfo/discuss
