705348 - Total CPU cost per silo report

Status: RESOLVED WONTFIX

(Release Engineering :: General, defect, P3)

Description

[Armen [:armenzg]]

I have been trying in bug 683734 to calculate how to distribute the snowleopard rev3 minis across pools.

I've tried to draw conclusions from pushes, waittimes and testruns reports.

I've realized that the correct way is the following:
* take every job that was run on certain silo (e.g. talos-r3-fed-XXX), add it up and remove the cost generated by certain jobs [1]

Once we have a total for each silo we can then choose the ones we want (in my case fedora, fedora64, leopard, win7 and winxp) and determine how to distribute the addition of new slaves.

We should be able to choose a starting date and an ending date.

We could refine this report by removing outliers but I would not want to overcomplicated this.

[1] For certain silos we run fuzzer on idle times. jetpack should be counted.

Comment 1

[Armen [:armenzg]]

Try jobs for the build side should be considered other set of silos win32-pool VS win32-trypool.
This is not the case for testing machines.

Comment 2

[Chris Cooper [:coop] (he/him)]

Armen: should we be talking to the Metrics team about this?

Comment 3

[Armen [:armenzg]]

(In reply to Chris Cooper [:coop] from comment #2)
> Armen: should we be talking to the Metrics team about this?

catlee, what do you think?

I think this is something we could do ourselves but this is probably something they can create for us from build.mozilla.org/builds.
I believe there are startime, endtime for each job and the slave that took each slave.

I would be happy if they took care of it :)

Comment 4

[Chris AtLee [:catlee]]

We certainly do have the data for this. Not sure if we have the time though :)

Comment 5

[John O'Duinn [:joduinn] (please use "needinfo?" flag)]

Found in triage.

Comment 6

[[e-mail] (:bhearsum)]

Silos are a Buildbot thing. Buildbot is dying.