Event Builder performance
- The server is SUN Ultra300 with 2 CPU (250 Mhz, 4MB Cache, UltraSPARC), 256 MB of memory , 4.2 GB 7200 RPM Wide Ultra SCSI disk.
- CPU_sec is a 1 sec execution on single 250 Mhz UltraSPARC CPU
- RT_sec is 1 sec of real time on the server
- The data was generated by event simulator running on the second CPU and sending data via UDP socket (as real ROC should do).
- The red KB/RT_sec curves are saturated due to the limited performance of the event simulator, the blue curves - due to the disk throughput.
- For event size between 8 and 16 KB the writing performance is 6-8 MB/sec, one CPU is completely busy. The difference between ‘simple’ and ‘standard’ TPhEvent is less than 30%. (Explanation: the main overhead is inside the streamer function). This performance expected to be higher with new disks (they have larger cache).
- For shorter events the performance is limited by maximum possible event rate (3300 and 1700 ev/sec)
- For longer events the CPU consuming is 3 times less for ‘simple’ TPhEvent.
- If we need better performance then the program should be made multithreaded to exploit other CPUs.