[Asterisk-Dev] sched.c::schedule function can be a bottleneck and is optimised backwards

[Kevin P. Fleming]

Steve Kann wrote:

> If you end up with thousands of schedulings in the first 100ms or so, 
> and then a thousand or so things spread over the next 30ms, then it 
> might make sense to split it up into two queues: a "short" queue, and a 
> "long" queue, or something like that.

That was going to be my suggestion :-)

Either create a separate scheduler context in chan_iax2 just for 
registrations, or have the context contain queues for stuff scheduled 
less 2000ms away vs. stuff scheduled longer.