Re: ovsdb scaling tests

Jamo Luhrsen <jluhrsen@...>

Thanks for this input Sharon. Are these issues in bugzilla, or maybe things
are already addressed in ovsdb itself?

Marcus, did you have any other data points for OVSDB?

All of these scenarios can (and should) be introduced in to CI so we can flush
out all the issues, and more importantly keep them from creeping back in.
I know there have been some test patches in the past trying to get us there,
but nothing finished yet.


On 08/28/2015 10:52 AM, Sharon Aicler wrote:
A few months ago I performed some scale tests on ovsdb Plugin in ODL with one OVS that was extremely loaded (around 1000 ports). I found three hot
spots that need to be attended (unless those points were already attended):

1) The OVSDB statistics column is being update very frequently, the ODL plugin is registering on all of the OVSDB tables & columns, hence it receive
all the OVSDB data every second or so.

2) The status column in each table is also updated very frequently, hence it should be extracted from the main query and have its own query so on
notification will get only this field new value and not the entire OVSDB data.

3) There was/is a MAX buffer size for the OVSDB data that prevented receiving large reply from the OVSDB, hence when there were more than 100 ports
defined on the OVS you would have got an empty reply.

After locally patching those on my code base I have not seen any other scaling issues.

One note, in a cluster setup ovsdb plugin should follow the "master" of the Topology shard, not sure it is doable today (or how to do it).


On 08/28/2015 10:35 AM, Luis Gomez wrote:
cc-ing Marcus, he is the one I know doing this test.

On Aug 28, 2015, at 2:26 AM, daya kamath <<mailto:daya_k@...>daya_k@...> wrote:

hi jamo and luis,

have we done any scale type tests with the ovs? if we have such data available, would appreciate if you can point me to it.


ovsdb-dev mailing list
Thanks, Sharon.

Join { to automatically receive all group messages.