[asterisk-dev] Something wron with Trunk

[Venefax]

3 days ago I upgraded a client who has 20.000 (twenty thousand) peers and was working with a 1.2 version since 3 years ago. I installed trunk and it core dumped 4 times in day and also it became irresponsive, for I could login with asterisk -r but calls would not get processed and a "restart now" did not work. I had to downgrade  to 1.4 and it has been stable for 24+ hours. So something is very wrong with Trunk. If a Digium-authorized developer wants to do some research on this issue, I can install trunk again and let it run, but definitely under the supervision of a developer. I believe that this is a unique opportunity to get a better Trunk, because there are not many systems with this kind of load --that we can manage. 
The architecture is like this: The call comes in, and I fire a Perl script against SQL Server, which determines if the call will go on and what carrier will terminate it. Basically, the database sends back an argument for the Dial application, which is executed after the script ends. The CDR is written to the same SQL database using cdr_odbc. I have a powerful reason to use Trunk, and it is that Trunk uses direct-sql, versus the prepare, execute and unprepared model used in 1.4. This model is crap, because it uses 3 network calls for each SQL statement, thus it is poorly scalable. 
Please contact me directly if somebody wants to debug this unique system. I believe it is the largest Vonage-model installation using a one-box-Asterisk in the country.