http://wiki.freeswitch.org/wiki/Performance_testing_and_configurations lists some performance bottlenecks, I'd like to check if they are still an issue... I recall some changes in how threads are run.
1) "libsofia only handles 1 thread per profile, so if that is your bottle neck use more profiles"Is this still true?
I saw Anthony mention in a ML post that there is a param for a separate thread for incoming registrations but I can't find it on the wiki or find the email.
2) "Reports of running more than a single instance of FreeSWITCH has helped."
If libsofia doesn't have a bottleneck (see #1), then what part of FS would? Each call get it's own thread. Why would this help?
3) "On a normal configuration, core.db is written to disk almost every second, generating hundreds of block-writes per second. To avoid this problem, turn /usr/local/freeswitch/db into an in-memory filesystem."
It seems the other alternative is:
"-nosql -- disable internal sql scoreboard"
What does -nosql disable? e.g. for a dialer trunk you don't need presence, track-calls.. can you still get an active call count from FS?
4) from
here "initial-event-threads
Number of event dispatch threads to allocate in the core. Default is 1.
If you see the WARNING "Create additional event dispatch thread" on a heavily loaded server, you could increase the number of threads to prevent the system from falling behind."
If I expect high cps/load, e.g. for dialer term, should I set this higher from the start?
Anything else I should know, other than "FreeSWITCH is free, so you can use that money you saved to get good hardware" ? I remembered that part!
-Avi Marcus
BestFone