Re: Todo list for the new talos machines

Where does baselining fall in with this?  When we first started the 
talos work there was a big interest in collecting the numbers going back 
a couple of years - which was done with some of the winxp blade 
machines.  While this remains on the goals list I haven't heard much 
about it.  Are we still interested in these numbers?  Is there something 
else to be found in them that hasn't been turned up by the first 
baselining attempt?

Bringing up this set up machines seems like a high priority (most 
everything is marked as priority #1) - yet I already have a full set of 
goals for Q4 for performance.  I know that we've had a lot of goals 
discussions and that everyone is probably tired of the topic, but I 
think that we are doing the same thing again where we set goals and then 
end up throwing them away a couple of weeks later.  Which doesn't bother 
me much as long as work that people are interested in is getting done, 
but does mean that we end up going through the "why didn't we hit our 
goals" talk in 3 months.

alice.


Stuart Parmenter wrote:
> vlad, rob, bhearsum and myself sat down this morning at FSOSS and went
> over what we needed to move forward with the new talos setup.
> 
> For each platform, I have suggested that we have 3 identical machines
> running at half speed (speedsteped down) with more cycles than we are
> currently using.  3 machines so that if one is acting up we can tell
> that the other two aren't.
> 
> Machines that we need:
> Priority | OS
> ---------+-------------------------
> #1       | XP
> #1       | Mac OSX 10.4
> #1       | Linux (Ubuntu)
> #2       | Mac OSX 10.5 (final)
> #3       | Vista (super duper edition?)
> 
> We would also like to have 1 non-chrome machine for each platform so
> that we can more easily identify chrome-only regressions.
> 
> (This is a total of 18 machines)
> 
> Todo:
> - Speedstep on mac
> - Speedstep on linux
> - talos GC patch (sicking)
> - increase cycles (15 or 20 -- whatever the magic number for lowest
> noise is)
> - image machines so that we can quickly deploy new ones
>