On Thu, Mar 5, 2015 at 4:45 AM, [email protected] <[email protected]>
wrote:

> On Thu, Mar 5, 2015 at 12:05 PM, Sven Van Caekenberghe <[email protected]>
> wrote:
> > Phil,
> >
> >> On 05 Mar 2015, at 11:44, [email protected] wrote:
> >>
> >> Images are starting slow, that's my experience. Here, a Pharo3 on
> >> Windows with quite a bunch of things in it takes six seconds to load
> >> on first launch and three after that.
> >>
> >> And that's on a i7 4770K clocked 3.85GHz with SSD drives.
> >>
> >> It takes less time to revive a VMWare VM :-(
> >
> > Maybe it is Windows ? Or certain extra stuff that you have in your image
> that does work at startup ? Because you should see sub second startup times.
> >
> > (This is on the smallest Digital Ocean instance, 1 virtual CPU, 512 Mb
> RAM, SSD)
> >
> > root@stfx:~/experiments/pharo4# curl get.pharo.org/40+vm | bash
> >   % Total    % Received % Xferd  Average Speed   Time    Time     Time
> Current
> >                                  Dload  Upload   Total   Spent    Left
> Speed
> > 100  2885  100  2885    0     0  17974      0 --:--:-- --:--:-- --:--:--
> 18031
> > Downloading the latest 40 Image:
> >     http://files.pharo.org/image/40/latest.zip
> > Pharo.image
> > Downloading the latest pharoVM:
> >         http://files.pharo.org/vm/pharo/linux/stable.zip
> > pharo-vm/pharo
> > Downloading PharoV10.sources:
> >         http://files.pharo.org/sources//PharoV10.sources.zip
> > Downloading PharoV20.sources:
> >         http://files.pharo.org/sources//PharoV20.sources.zip
> > Downloading PharoV30.sources:
> >         http://files.pharo.org/sources//PharoV30.sources.zip
> > Creating starter scripts pharo and pharo-ui
> >
> > root@stfx:~/experiments/pharo4# ./pharo Pharo.image printVersion
> > [version] 4.0 #40535
> >
> > root@stfx:~/experiments/pharo4# ls -lah Pharo.image
> > -rw-rw-r-- 1 root root 22M Mar  4 14:51 Pharo.image
> >
> > root@stfx:~/experiments/pharo4# time ./pharo Pharo.image eval '100
> factorial'
> >
> 93326215443944152681699238856266700490715968264381621468592963895217599993229915608941463976156518286253697920827223758251185210916864000000000000000000000000
> >
> > real    0m0.386s
> > user    0m0.184s
> > sys     0m0.028s
> >
> > root@stfx:~/experiments/pharo4# time ./pharo Pharo.image eval '100
> factorial'
> >
> 93326215443944152681699238856266700490715968264381621468592963895217599993229915608941463976156518286253697920827223758251185210916864000000000000000000000000
> >
> > real    0m0.422s
> > user    0m0.192s
> > sys     0m0.032s
> >
> > root@stfx:~/experiments/pharo4# time ./pharo Pharo.image eval '100
> factorial'
> >
> 93326215443944152681699238856266700490715968264381621468592963895217599993229915608941463976156518286253697920827223758251185210916864000000000000000000000000
> >
> > real    0m0.863s
> > user    0m0.404s
> > sys     0m0.064s
> >
> > It is the latest (smaller) image, but that should not make much
> difference.
>
> On my CentOS Linux, headless tests are fast as well.
>
> Granted, my Windows image was super large. 300Megs or something.
> I cleaned the Monticello things with flush cache, which removed 41 megs.
>
> I can't test the CLI version on Windows, as I do get a window opening.
>
> There are a ton of windows and morphs open in there and I've run a
> couple of experiments.
>
> Now, the Linux image was large too. 200Megs or so.
> And the window showed immediately, then it took a second and a half or
> so to show the images back.
> There is a black background before.
> Maybe the system is processing its startUp list, unhibernating forms,
> I don't know what's going on but it takes that 1.5 secs.
>
> The image gets large because of Monticello caches, and TWM (Tiling
> Window Manager) keeping a handle on browsers etc over time. So, I
> cleaned the caches, closed TWM, gc'ed a time or two and then got the
> image back to 90 megs.
>
> Then the image opens in like 0.5 sec. It is hard to measure as what I
> want is the time between issuing the pharo-ui XYZ.image and the fact
> that I can actually see the UI.
>
> How would one do that?
>
> Sorry for the rant but generally speaking, the Pharo UI seems slower
> than the Squeak 4.5 ui.
> Or the EToys UI for that matter.
> I am using those for little educational programs and they feel faster.
> I'd like to know why. Both systems have morphic. Is it because of
> aliased fonts? Squeak 4.5 also has that.
>
> Is there any way to monitor what's going on during image startup?
>

I modifier the AndreasSystemProfiler in the Squeak code base at Cadence to
allow profiling across snapshot; you see both save and startup profiles.
That's not showing e.g. disc activity from VM reads, but it does help track
down tike spent in the SMalltalk start-up.  IIRC this is in te Saueak code
base.  But if you're interested I can dig it out.  The changes are minor;
to do with correctly offsetting delays across snapshot.


> Phil
>
>
>
>
>
> >
> > Sven
> >
> >
> > --
> > Sven Van Caekenberghe
> > Proudly supporting Pharo
> > http://pharo.org
> > http://association.pharo.org
> > http://consortium.pharo.org
> >
> >
> >
> >
>
>


-- 
best,
Eliot

Reply via email to