On 5 Jul 2017, at 19:03, Henrik Sperre Johansen <henrik.s.johansen@veloxit.no> wrote:
Sven Van Caekenberghe-2 wrote
On 27 Jun 2017, at 21:15, Sven Van Caekenberghe <
sven@
> wrote:
On 27 Jun 2017, at 21:00, Esteban A. Maringolo <
emaringolo@
> wrote:
Interesting experiment.
It would be good to know how fast does the image boot and start execution in the context of a service billed at the sub second scale.
Sadly I have the impression we lost some speed:
~/pharo64$ ./pharo Pharo.image printVersion [version] 6.0 #60499
~/pharo64$ time ./pharo Pharo.image eval '42 factorial' 1405006117752879898543142606244511569936384000000000
real 0m1.276s user 0m0.985s sys 0m0.171s
while before
~/pharo/build$ ../bin/pharo Pharo.image printVersion [version] 4.0 #40620
$ time ../bin/pharo Pharo.image eval '42 factorial' 1405006117752879898543142606244511569936384000000000
real 0m0.526s user 0m0.434s sys 0m0.045s
But it is really hard to find the cause.
The good news is that the Pharo 7 minimal core is way faster
$ time ./pharo ~/Downloads/Pharo7.0-core-32bit-b649ac6.image eval "42 factorial" 1405006117752879898543142606244511569936384000000000
real 0m0.262s user 0m0.132s sys 0m0.062s
Which is good of course !
So I guess, most of the additional time is either spent in #startUp methods not present in the minimal image (), or in vm reading the larger sized image? While doing the same in Pharo4 would probably be a pain, it could be interesting to hack ClassSessionHandler in Pharo7 to log time spent in each, and compare the output difference between minimal/full image...
Yes, there should (optionally) be some log with timing info explaining which startup actions are executed and how long each takes. Maybe the log could just be kept in memory in the image.
-- View this message in context: http://forum.world.st/Are-we-able-to-run-pharo-images-in-a-read-only-environ... Sent from the Pharo Smalltalk Users mailing list archive at Nabble.com.