Q10020 Accuracy of STATISTICS report
How accurate is the STATISTICS
command for benchmarking?
The results of the STATISTICS
timing start out no more precise than the time resolution that DAPL processing tasks see. This basic interval is the 50 nS time interval between cycles of the precision oscillator/timer. The number of samples processed is determined by the more stable sampling clock, but the statistics processing collects data at time intervals based on the processor clock. Because of the independent clocks, there is some uncertainty in how the two are coordinated for the STATISTICS
collection. There is some delay for statistics update processing. The process is started and stopped by the interpreter task, with additional task scheduling uncertainty depending on high-priority critical processes.
All of this says that the STATISTICS
gives useful information about task loading, resource conflicts, excessive latency and such over time, but it isn't very precise about short-term phenomena. There will be some variability in the results.
L11657