[nested] inclusion of u-s-c as system comp not getting system load zero as quick as before

[kevin gunn]

So our man ogra did some work to promote nested mir into touch images.
however, noticed that the wait periods for system settle seem to be getting longer

for example
http://ci.ubuntu.com/smokeng/trusty/touch/mako/176:20140212:20140115.1/6550/ubuntu_calculator_app/

and this is top output fo the system settle
https://jenkins.qa.ubuntu.com/job/trusty-touch-mako-smoke daily/33/artifact/clientlogs/ubuntu_calculator_app/topafter.log/*view*/

[Oliver Grawert (ogra)]

just some background info here:

the mentioned test goes through up to 10 iterations of 5 loops of top, with a 6 second delay between each until it either falls out (fail) or sees that the system has gone to 97.5% idle or above

since we add a new layer to the stack with nested Mir mode, i actually think we can expect the system to take a little longer to settle and would also propose to adjust the test parameters accordingly ...

even asac agrees with that view, but nontheless he wants a detailed research and explanation of the issue before he approves this step.

[Daniel van Vugt (vanvugt)]

I'm not sure this will be easy to resolve. Nesting obviously doubles the number of Mir servers you're running. And a single Mir server is not quite light-weight yet (see bug 1236508).

Last time I analysed CPU bottlenecks, the biggest problem was the Mir protocol logic (protobuf) which will be a problem doubled if you're nesting.