Le 5 mars 2015 22:54, "Eliot Miranda" <[email protected]> a écrit :
>
>
>
> 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.

Definitely!

Phil
>
>>
>> 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