Notes on Debugging Performance Issues
The following are some notes on debugging performance issues:
If the amount of time taken to stop and start SBRC is more than two minutes, there might be too many sessions to be sustained through the current I/O bandwidth. You might also need to shard or move to cluster(s), or put the data files on a Fibre Channel storage device or similar devices. Also examine the CPU and I/O utilization at startup for added insight.
The accounting and proxy thread usage is a function of downstream latency. Many threads will have low CPU utilization, since most of the time is spent waiting. Each thread has some overhead and takes a finite amount of memory. If the system requires more threads, it is trying to take in more packets than the downstream can receive, in which case you should consider if proxy spooling is a better solution.
Accounting-ON and Accounting-OFF are work-intensive operations, as you can configure SBRC to generate AutoStops to be sent downstream or against the SSR. This generates a great deal of work for SBRC, the SSR, and downstream proxy targets to handle one RADIUS packet’s worth of input. Taking this into consideration, the system may hit limits while processing these operations if there are many sessions to be terminated.
See Accounting-Off or Accounting-On with AutoStop Enabled for more information.