[asterisk-dev] invasive fixes to chan_iax2 in 1.4

[Russell Bryant]

Loic Didelot wrote:
> in this case it was a bug related to iax registrations. Try to upgrade
> to asterisk 1.4 and you can far more then 7k registrations.
> 
> Just for your information:
> I use mysql but not on the same machine then asterisk but on the same
> lan. Mysql is behind LVS.

A *great* side benefit of converting this code over to astobj2 is that the peers
and users are now stored in a hash table instead of a single linked list.  The
way the code works during a registration, the peer has to be looked up in this
container multiple times.  When using a lot of peers, You should be able to get
much higher results in a registration load test with the iax_refcount branch,
and also 1.4 once the changes are merged in.

On another note, I just realized a change in behavior that these changes
introduce.  Previously, iterating peers and users to find a match was done in
the same order that the entries were put in configuration.  However, they are no
longer stored in any guaranteed order.  If there is any setup that depends on
the order in which peers and users get matched against, it will no longer work.
 I don't *think* this is worth considering as a problem, but please speak up if
you feel otherwise.

-- 
Russell Bryant
Software Engineer
Digium, Inc.