Ok, I have enough for a benchmark, just short of being able to provide a
working implementation.
I actually ran out of memory with the Stem implementation, it seems to use
a lot of additional memory when running the event loop. I had to reduce
the number of handles to 20000.
Otherwise, it looks quite good (this is the synthetic benchmark that
doesn't test event loop performance, just overheads):
name watchers bytes create invoke
EV/EV 100000 244 0.54 0.45
Perl/Any 100000 513 4.61 0.87
Stem/Any 20000 1159 31.51 9.12
Event/Any 16000 936 38.80 31.91
POE/Any 2000 6658 108.61 726.39
It is only 10 times slower than anyevents pure perl backend, but three
times faster than Event.
For the large server benchmark, it still looks very good:
name sockets create request
EV 20000 68.51 10.65
Perl 20000 73.58 112.90
Event 20000 199.29 252.45
Stem 20000 128.11 660.92
Glib 20000 632.13 1818.40
POE 20000 344.56 12154.89
Only 6 times slower than anyevents pure perl backend, but faster than Glib
(ok, glib clearly has issues :)
So it is actually quite good, and quite usable, except for the out of
memory issue.
(This is without the multiple watchers per fd/combo workaround, btw.,
which would only affect create speed).
It hurts a bit to come so far and then to only fail because stem has no
one_iteration-like functionality, though (I would go with busy waiting, if
that were implementable, but thats not reasonable to implement either).
--
The choice of a Deliantra, the free code+content MORPG
-----==- _GNU_ http://www.deliantra.net
----==-- _ generation
---==---(_)__ __ ____ __ Marc Lehmann
--==---/ / _ \/ // /\ \/ / [EMAIL PROTECTED]
-=====/_/_//_/\_,_/ /_/\_\
