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
