[Stackless] Bug in stackless.tasklet C API?

Andrew Francis andrewfr_ice at yahoo.com
Thu Apr 24 16:17:21 CEST 2008


Hello Lucas:

>The thing that might prevent me from actually using
>it, is that I'm having a very hard time finding a
>solution to hook up stackless to network sockets, in
a >way that will reliably deal with thousands if not
much >more, of connections. Richard's stacklessocket,
>Phoenix's attempt at socketlibevent socket, and the
>recent Twisted network >hookups all don't feel like
>they'd hold up when things >start to scale up.

I suspect almost any application will have problems
handling thousands of simultaneous connections -
without a combination of hardware, operating system,
software, and network topology tricks.

That said, I pick ease of modeling the problem domain
over scalability any day. Stackless Python is great
for modeling.

>Mind you, this is very much a feeling, so if anybody
is actually using any of these in a similar
environment, with loads of connections, I'd love to
hear about it.

(Oops I knew I forgot something....)

Well a few weeks ago I posted a simplified version of
a Toy BPEL processor that illustrates all the
techniques I discussed at PyCon 2008. The server was
not built for speed - the server has no optimizations
(i.e., channel preferences, send_sequences) and also
has a race condition that should result in a 404
(really should be 500). 

My biggest problem to date is that I don't have a
machine big enough to pound the example (i.e., I ran
out of file descriptors on Windows XP).

Cheers,
Andrew



      ____________________________________________________________________________________
Be a better friend, newshound, and 
know-it-all with Yahoo! Mobile.  Try it now.  http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ




More information about the Stackless mailing list