Measuring Sub-microsecond Speeds

[gregory...@nist.gov]

Hello PFT community,

We have been increasingly receiving algorithms that operate at speeds faster than previously anticipated. To better facilitate analysis of template comparison speeds, time values for this operation will now be reported in microseconds instead of milliseconds.

Some algorithms have template comparisons speeds less than one microsecond. Unfortunately, PFT III's measurement code only recorded durations in microsecond resolution. The clock on our timing test CPU is capable of accurate measurements to the nanosecond resolution. As such, we have updated our testing code going forward and have additionally re-run all timing tests for all algorithms adhering to PFT III API v1.1.0. This has introduced slight changes in Section 2 ("Timing Sample Dataset") of the report cards for these algorithms, but values are all nearly the same as previously published. No other analysis has changed.

We have additionally released machine-readable versions of time values from report cards as well as a combined DET featuring the latest algorithms for PFT III-specific datasets.

Thanks,
-Greg (@nist.gov)