>Hello Rob,
>
>It is a known problem. I don't have the PMR & Bug number, it is
>somehow alleviated by TL05. Shortly put, there is a memory leak
>when calling a specific OS function (crypt) and it manifests with all
>long running programs calling it (Informix/Db2/...).
>
>You can test it for yourself (as we did), generating connections
>to the database (doing nothing and exiting as soon as possible,
>as many connections you can do). After about 1.3 Mil connections
>on pre-TL05 and after about 4.2 Mil connections on post-TL05
>you hit this limit.
>
>We are also interested in a more permanent solution - as it is now,
>we still have to bounce the engine once 3-4 weeks of activity.
>If you get more information about this, please notify me also
>(or post in the newsgroup).
>
>Regards,
>Bogdan BOTEZ.
Bogdan kindly emailed me to point us in the right direction. We
contacted Informix support again and there doesn't seem to be a
solution to this problem as of yet. The cause of the problem seems to
be related to the total amount of CPU time spent on the msc-VP (last
column of:onstat -g glo|grep " msc "|tail -1) so when it gets above a
certain level then the problem will start occurring. Once identified
(write a simple script that opens up and closes a lot of connections to
a database on the server), this could be used as a metric to decide
when to bounce the instance (until the AIX blokes finally get around to
fix this).
Thank you all for your input!
RoB