[Asterisk-Dev] Profiling results: Asterisk processing 5000 concurrent IAX registrations

[steve at daviesfam.org]

On Mon, 31 Oct 2005, Steve Kann wrote:

> Kevin P. Fleming wrote:
> 
> > steve at daviesfam.org wrote:
> >
> >> In this example, looks like we'd probably score significantly by
> >> down/upshifting peer names or whatever and then using a standard 
> >> str(n)cmp rather than strcasecmp.
> >
> >
> > There are actually many places in Asterisk where that sort of thing 
> > will be beneficial, and I've done quite a bit of work in that area for 
> > post-1.2 merging... the same is true in the dialplan, only there the 
> > problem is even worse due to the multiple searches that happen for 
> > each step.
> 
> 
> Would it make a big difference to just use strcmp instead of strcasecmp? 
> My guess is that it would improve things by at most 25% or so, but 
> probably less (like 10%).  That's just my gut, though..
> 
> Obviously, you need to move from a linear search over all the peers 
> O(n), which is does n times (so the whole registration system is 
> O(n^2)), to some kind of indexed search..

Yeah - agreed.  Kevin was talking at some point at Astricon (missed you, 
BTW) about new structures for all the many lists in Asterisk.

I'm happy to work on that, but I got the impression something was in the 
works already.

I'd expect strcmp to be a lot faster that strcasecmp. by the way, because 
there are the special and fast "rep cmp" type instructions on the 80x86.

Steve