throttle mac mini XP speed down to something slower

RESOLVED FIXED

Status

Release Engineering
General
P2
normal
RESOLVED FIXED
11 years ago
5 years ago

People

(Reporter: vlad, Assigned: bhearsum)

Tracking

Firefox Tracking Flags

(Not tracked)

Details

Here's some instructions for throttling the CPU speed of the mac minis down:

1) Download and install SpeedswitchXP -- http://www.diefer.de/speedswitchxp/

2) Run from start menu; it should start up automatically on restart.  Hit Detect max cpu speed just to get that number on there.

3) Go into Control Panel -> Power Settings.  Set the power mode to Speedswitch Controlled.

4) Back in Speedswitch, CPU speed on External Power should say "Max. Performance".

5) Open Options.  Check "Read CPU speed [Ask Windows about speed]" and "Show history diagram".  Set the scaling to 5x or something.  Close options.

6) Watch the cpu speed graph draw for a bit.

7) Go into Options again, go to User Policy tab, and under "On external power", in Forced throttle, set an appropriate number.  50% produces 800MHz on my mini.  It can go lower; there is a lower bound, however.  I can get my mini down to 233MHz.  Set to whatever value.

At this point you should do something like open up a window and drag it around the desktop really fast to generate some CPU load, and make sure the number doesn't go up above the now-lower line on the graph.

Two things:

- We can do something similar on Linux if necessary; we can probably do something on OSX as well, but I'm not sure how.

- SpeedswitchXP is GPL and comes with source; someone can make a command line tool that does all this that we can just run before perf runs (and possibly run after to restore the speed for non-perf-run parts).
(Assignee)

Comment 1

11 years ago
I'm going to install this on a mini today or tomorrow and do some perf runs.
Assignee: nobody → bhearsum
(Assignee)

Updated

11 years ago
Status: NEW → ASSIGNED
Cool, would be good to keep at least one mini running at full speed so we have an idea of what perf is like on an unthrottled machine.  But setting this up on a separate mini initially sounds like a good idea.
(Assignee)

Comment 3

11 years ago
I didn't get to this as soon as I thought. It's on my list for tomorrow.
Priority: -- → P2
(Assignee)

Comment 4

11 years ago
I've got this running on an XP machine now. Here's the results so far at 50% speed (~800mhz) compared to full speed: http://graphs-stage.mozilla.org/#spst=range&spstart=1187981266&spend=1188515740&bpst=cursor&bpstart=1187981266&bpend=1188515740&m1tid=5204&m1bl=0&m1avg=0&m2tid=280&m2bl=0&m2avg=0

Comment 5

11 years ago
ugh, can you please do the slow numbers on another machine?  the one you're doing them on were being used to compare against the 1.8 box :(
That looks great, I'm glad that it worked!  Stuart, aren't the numbers from mini02 the same?  Or, we should slow down the 1.8 box as well, since the numbers seem stable.
(Assignee)

Comment 7

11 years ago
I've been using xp01 as test machine for the speedswitch stuff. I'm using xp02/03 as the production machines. If you're satisfied with how speedswitch is working I will go ahead and set it up on these two machines.

I'm also planning to move these to graphs.m.o. How does that sound to everyone?
Group: mozillaorgconfidential
As a note, no machines will be moved to either the production tinderbox waterfall or the production graph server without a separate bug filed to request it.  That discussion does not belong in this bug.
(Assignee)

Comment 9

11 years ago
We have two throttled windows XP builders (qm-mini-xp01/05). I'm going to mark this FIXED. If/when we want to throttled Linux and Mac specific bugs can be filed about it.
Status: ASSIGNED → RESOLVED
Last Resolved: 11 years ago
Resolution: --- → FIXED
Group: mozilla-confidential
Mass move of Core:Testing bugs to mozilla.org:ReleaseEngineering. Filter on RelEngMassMove to ignore.
Component: Testing → Release Engineering
Product: Core → mozilla.org
QA Contact: testing → release
Version: Trunk → other
Product: mozilla.org → Release Engineering
You need to log in before you can comment on or make changes to this bug.