"dev" <dev-boun...@openvswitch.org> wrote on 05/03/2016 10:45:46 PM:
> There are basically two possible paths here. One path is to enhance > OVSDB. The other is to switch to a different distributed database. Of > course, in the latter case the question is "which one?" Until recently, > we weren't seeing much performance or availability pressure on OVSDB, so > it made sense to stick with what already worked. Now, we're starting to > understand the requirements and the bottlenecks better, so it may be > time to consider what distributed database is right for OVN. Opinions > are welcome! > I am strongly support the option to switch to a different distributed database for the SB DB. It will allow us to focus in networking in OVN and get the benefits of a mature Open Source database (Scalability HA, security...). I think that we should summarize the requirements of OVN SB DB and then maybe we can build an abstraction layer in the client side (something like the ovsdb-idl) that will allow us to work with different DB implementations. While I am writing the conditional monitoring implementation in the ovsdb-server, I thought that we need a message queue layer that implements a generic publish subscribe interface between the ovn-controllers and the SB DB itself to ease the scalability of some of the tables in SB DB (Port_Bindning, Logical_Flow..). > On Fri, Mar 04, 2016 at 12:39:54PM -0500, Russell Bryant wrote: > > There's a lot of work happening to improve ovsdb performance (both on the > > client and server sides). There's testing happening in multiple > > environments (physical and simulated) in the hundreds-of-hypervisors > > range. Interestingly, most of the bottlenecks we're exposing are on the > > client side. > > > > We have some docs about the current HA story with OpenStack here: > > > > http://docs.openstack.org/developer/networking-ovn/faq.html > > > > Ben has mentioned that he might pick up the distributed ovsdb-server work, > > which is important to get a much better HA story (or I suspect we'll have > > to replace ovsdb). I'll let him comment further on intentions and status, > > though. > > > > On Fri, Mar 4, 2016 at 10:14 AM, Dan Mihai Dumitriu <dm...@cornell.edu> > > wrote: > > > > > Hi Ben, > > > > > > What's the current thinking around the OVSDB HA and scale solution? > > > Needless to say, the single SB DB to which all ovn-controllers connect > > > could be a liability in various production scenarios. > > > > > > Cheers, > > > Dan > > > On Mar 4, 2016 00:48, "Ben Pfaff" <b...@ovn.org> wrote: > > > > > > > Here's my OVN report for the week, since I'll be in a meeting during the > > > > IRC session today. > > > > > > > > The "continuation" feature needed as a basis for others is in. Justin > > > > is reviewing the ARP support patches. > > > > > > > > I've spent most of the week in meetings, so there's been minimal > > > > progress. I'm currently working on some debugging support patches. > > > > After that, I'm going to work on OVSDB. I'm also interested in > > > > improving ovn-controller performance. > > > > _______________________________________________ > > > > dev mailing list > > > > dev@openvswitch.org > > > > http://openvswitch.org/mailman/listinfo/dev > > > > > > > _______________________________________________ > > > dev mailing list > > > dev@openvswitch.org > > > http://openvswitch.org/mailman/listinfo/dev > > > > > > > > > > > -- > > Russell Bryant > _______________________________________________ > dev mailing list > dev@openvswitch.org > http://openvswitch.org/mailman/listinfo/dev _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
