Yes. They are probably creating a session file at every request. On Tuesday, 25 September 2012 10:16:44 UTC-5, Michele Comitini wrote: > > First try should be adding a session.forget() and see what happens... > > N.B. The benchmark in question uses really 4 concurrent processes > through uwsgi, so suspect number #1 is a file lock. > > Correctly the test was setup on two multicore machines one with the > server one with the client, so real concurrency comes into play. > > mic > > > 2012/9/25 Anthony <abas...@gmail.com <javascript:>>: > > web2py source is here: > > https://bitbucket.org/akorn/helloworld/src/145cbdf4f995/web2py > > make file is here: > > https://bitbucket.org/akorn/helloworld/src/145cbdf4f995/Makefile > > > > > > On Tuesday, September 25, 2012 10:02:40 AM UTC-4, Massimo Di Pierro > wrote: > >> > >> I agree we should try reproduce those benchmarks becomes something is > >> clearly very wrong. > >> I cannot find the code used for those benchmarks, so I added a comment > >> asking for it. > >> > >> > >> On Tuesday, 25 September 2012 07:59:13 UTC-5, Jose C wrote: > >>> > >>> Hi Massimo, > >>> > >>> I too agree that benchmarks, like statistics, can be very deceptive. > >>> > >>> The point is comparing just 2 of the frameworks that I'm personally > >>> interested in (and I would have imagined had similar startup > overheads), > >>> i.e. web2py and django, you see web2py getting 686 requests compared > to > >>> django's 15,346! That's a massive difference and like Michele's > comment, I > >>> wonder if there is something that can be learnt from this and some > >>> optimization performed that might help with future versions? The > numbers > >>> certainly look bad for any new person going through the process of > choosing > >>> a framework to start with. > >>> > >>> On the memory leak issue, the author says he hit it running the simple > >>> "hello world" script test. I imagine he's not creating a class with a > self > >>> reference as you mentioned for his simple test. > >>> > >>> Perhaps one of the devs could try simulate the test (the author seems > >>> that have released all the test code and setup scripts) and see > whether the > >>> memory leak issue is indeed present. > >>> > >>> P.S. I do realize that even django doesn't have sessions enabled by > >>> default and wouldn't be surprised if that factor alone accounts for > the > >>> difference. A person selecting a framework up front won't know that > though. > >>> Perhaps Massimo should point it out in the author's blog comments, > >>> specifically all the setup work being done by web2py to make the > framework > >>> real-world usable. > > > > -- > > > > > > >
--