On Mon, Sep 10, 2012 at 1= 1:27 PM, Robert Klemme <shortcutter@googlemail.com>= wrote:

On Mon, Sep 10, 2012 at 10= :16 PM, jason marshall <jdmarsha= ll@gmail.com> wrote:
> We have a Rails app running on Ruby 1.9.3, and a under Windows we'= re seeing
> a 20x time penalty in our RSpec tests.

And the hardware is comparable?

Yes, = good question. =A0On the same model hardware, it's 20x initial, 4.5x cu= rrent. (2000, 440, vs <100 seconds)

=A0

> I'm fairly new to Ruby, but not to performance analysis, so I'= ve volunteered
> to follow up on this problem. =A0If I want to file/find a bug to assoc= iate
> with this problem, what sort of data should I be gathering, and how?

I have zero experience with Rails profiling, but maybe using the std<= br> lib profiler can give you a first data point.

Yeah my readings from Google say it's not easy to do what I need= to do. =A0I'm hoping someone could talk me down from the ledge and poi= nt me to a baked solution.

=A0

> Background/random data:
>
> We have 800+ tests, and on a good day they all run in a minute. =A0Som= e do no
> I/O (pure 'unit' test), most do light I/O (usually a couple cr= eate+reads to
> MongoDB or a render partial). =A0Only a few do any heavy lifting.
>
> I don't see any glaring hotspots. =A0Most of the tests seem to tak= e several
> times longer to run. =A0Dots just accumulate slower.

Just to exclude an obvious issue: do you have an AV engine up and
running on the Windows box? =A0If yes, excluding some directories might
help. =A0Some engines then still slow down operation so completely
switching off for the test might also be an option to determine
whether it's AV.

I'll entertain= the question, but let's take it as a given that turning off AV on Wind= ows is a very, very unwise action.=A0

Those number= s reflect my project directory already being excluded. =A0I'm not sure = I'm comfortable having Ruby whitelisted, though it's an interesting= data point: =A0

With Ruby exclusion, Regular is 1490 seconds, patched i= s 400 seconds. =A0

To exclude something else: is there concurrent activity that eats CPU
and / or IO bandwidth?

Not much. =A0Or = at least, there's not a lot of jitter in the numbers. =A0I just have Ru= byMine and a web browser (low CPU) open.

=A0

Kind regards

robert

--
remember.guy do |as, often| as.you_can - without end
http://blo= g.rubybestpractices.com/

As a mostly unrelated anecdote, the Su= bversion team had conversations like this for a long time, and resisted mak= ing changes for a number of years. =A0In the 1.7 release they made some pre= tty substantial changes to how they use the filesystem, and ended up with s= omething that was both simpler (for users) and happened to work a hell of a= lot faster on Windows.

It sounds like you guys are on that path at least. =A0

--
- Jason