Online Level-3: Performance

Performance Tests

For the purchase of the Run II startup in March 2001 we have performed several performance tests which insure we are buying the most effective system for the Level-3 processing. These tests have been performed on September 25, 2000. The setup is described here and takes into account the particularities of the Level3 trigger farm application. More tests specific for the Level3 filter are also available here.

From the results of the CPU and memory and the filter tests we can conclude the following

• All hardware tested is compatible with the Level3 control and filter code. No particularities were observed and everything worked out of the box.
• Memory starvation is not a problem at all for the Processor Nodes. The machines with 128MB performs as well as the machines with 512MB. This might not be true for the executable which has memory leaks since we tested only a relatively short sample of 200 events (filter crashes after 200 events).
• Increased clock speed of the processors up to 600 MHz scales almost exactly. For the machine with 733 MHz: 6% of the expected performance increase is not observed. These results are consistent with tests performed by other people.
• Increased memory access speed 133 MHz versus 100 MHz cannot be converted into an increase in performance.
• The level3 control framework (including reformatter) takes a negligible amount of CPU which is what we hoped for.