[Chandler-dev] No more perf regression bugs?

Bryan Stearns stearns at osafoundation.org
Mon Apr 16 17:35:36 PDT 2007


Heikki,

I do think our performance is important, and that it's worthwhile to 
have a set of metrics to shoot for, and to have times (like right now) 
where we focus on those problems. However, I don't think 
performance-regression bugs targeting a particular commit are useful or 
positive for me as a developer:

- Performance analysis works better when all the changes affecting a 
metric are in place: we can analyze the whole chain, not just one piece 
at a time.
- It's not like we can just back out the commit and discard the feature 
requirement. Most of the time, it's not possible to implement new 
features without some performance cost.
- It's *really* demoralizing to work hard on a feature, then have a 
performance-regression bug filed against it a day or so later (usually 
after you've started to dig into the next feature).

I also have trouble with our performance-monitoring mechanisms: many of 
the measurements vary widely, even when run with the same version of the 
code: here's 19 runs against a single revision on a single platform, and 
the standard deviation is 1/3 of the average time!

http://builds.osafoundation.org/perf_data/detail_20070415.html#creating_a_new_event_in_the_cal_view_after_large_data_import.double_click_in_the_calendar_view

That's a weekend day: on a weekday, or a slower platform, there may be 
only one perf run of each revision (or none at all). Because of this, 
the graphs cover too short a period to reliably see the effect of a 
single commit.

This means that the graphs are really only indicators of an area where 
investigation is necessary. I'm still working out how best to do that 
investigation... for me, so far, it's been best to start with profiling 
part of the code being run by the performance test; I've learned to 
avoid profiling the voodoo idles and sleeps that the functional tests 
contain: they only throw off the measurements. Once there's a profile, I 
can look for a method or functional area that's too slow or called too 
often, and rewrite as necessary. To do this, I don't look back at 
whatever revisions happened in the past.

(While we're on the subject: I also don't like the way we state our the 
performance targets: If we say that 1 second is "acceptable", but the 
"goal" is .1 seconds, I'm going to stop looking at a problem once it 
reaches "acceptable" and switch to another problem, and won't try to 
improve the first one further until all the other metrics are at the 
"acceptable" level -- and probably not until after all other bugs are 
fixed, too, which hasn't happened yet. I'd be happier if the table on 
the tbox page used a shade of green once a measurement got to 
"acceptable", and a brighter shade if it got to the "goal": the table 
would make us look a lot less screwed than the red and orange mess there 
now, which makes it look like we're making no progress at all.)

...Bryan

Heikki Toivonen wrote:
> I've been under the impression that we cared about performance
> regressions, although not (yet?) enough to back out a change that caused
> one. My thought was that we'd get the bugs filed when they happened, and
> analyzed the changes later to determine what caused the regression. I
> also thought that it would be easier to determine the cause of the
> regression (and fixing it) by working with the change information rather
> than just doing regular performance work (start with profile etc.).
>
> However, some engineers at least don't want to work with this past
> change information, and won't be looking at performance bugs at all, as
> far as I understand.
>
> I'd like to know if everyone feels this way. If yes, then I can stop
> filing performance regression bugs. I'd probably be able to simplify the
> performance reporting tools quite a bit if we didn't care about
> regressions (like maybe only reporting the absolute test results on
> tbox, and nuking all other results).
>
> If some feel perf regression bugs are usable, then I have a question
> about what to do with perf regression bugs that are caused by an
> engineer who does not want to look at performance regression bugs.
>
> I am somewhat annoyed by the situation, mostly because I'd like to avoid
> doing work that is considered useless. I do understand people have
> different styles of approaching problems.
>   



More information about the chandler-dev mailing list