We are hosting a WCF server farm. We currently have 12 app pools, but whenever we increase the threads per pool, we end up getting 503 errors. The service stops responding.
How does one tell what is the best number of threads per app pool?
Take a memory dump of the application pool in its broken state, and get the app developers to debug it.
You may need to use ProcDump from sysinternals, and OrphanActionExe and related parameters to trigger a dump automatically when the pool breaks.
If the 503s are immediate, you may find the app pool is failing immediately, which would be a slightly different troubleshooting path.