722537 - Add mozconfigs for 10.7 builds to mozilla-central

Status: RESOLVED FIXED

(Release Engineering :: General, defect)

Description

[John Ford [:jhford] CET/CEST Berlin Time]

In preparation for our move to 10.7 based machines, I'd like to have mozconfigs checked into the tree for these machines to use.  I don't want to use the 10.6 mozconfigs because I know there will be areas where these will diverge from the 10.6 machines.

Strategy for moving them to the 'normal' macosx could be:

1) discontinue builds on 10.6 that use 'macosx'
2) land patch to overwrite 'macosx' mozconfigs with 'macosx-lion' content
3) land patch to change the location buildbot code uses for the mozconfig to point it to the 'macosx' location
4) land patch to remove 'macosx-lion' from m-c

Comment 1

[John Ford [:jhford] CET/CEST Berlin Time]

Attachment: mozconfig changes

For now, the only change needed is -j4 being replaced by -j12

Comment 2

[Daniel C]

Did you do additional tests to determine that -j12 was the best choice? It seems from bug 720788 that quad-core i7s with HDDs are being ordered, which is not a configuration that was tested as part of http://blog.johnford.org/new-mac-builders-ssds-j-settings/. In addition to that, no values of -j were tested between 4 and 12 other than 8.

Comment 3

[John Ford [:jhford] CET/CEST Berlin Time]

(In reply to Daniel Cater from comment #2)
> Did you do additional tests to determine that -j12 was the best choice? It
> seems from bug 720788 that quad-core i7s with HDDs are being ordered, which
> is not a configuration that was tested as part of
> http://blog.johnford.org/new-mac-builders-ssds-j-settings/. In addition to
> that, no values of -j were tested between 4 and 12 other than 8.

The machine weres purchased with 2x500GB hard drives and are set up in a Raid 0 configuration as that was shown to be around 1.5 minutes (https://bugzilla.mozilla.org/show_bug.cgi?id=717102#c6).

The difference between -j8 and -j12 is minimal.  Getting the machines up at -j12 is much more important than tuning a single parameter.  Getting these machines up at -j12 will cut our build times from >4 hours to around 1h push to builds available.  I think that getting a >3 hour improvement into production is more important than benchmarking one tuning parameter.

If you have an identical spec mac mini and the cycles to test this, I'd be interested in your results!  We are running Core i7 2.0GHz quad, 8GB ram, 2x500GB in Raid0.  Because of all the variables, any tests done on non-identical hardware wouldn't help.