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
