--- Begin Message ---A company asked me for data on interrupt response times in FreeBSD and I've done a setup in my lab to measure it.I have completed the first half of the task: characterising the jitter on the response time, and the data are presented for all to see here. The second half of the task is more tricky: measuring the delay from the external stimuli until we execute the first statement in the interrupt handler, but if I can crack that nut data will be forthcoming on that as well. These data are for the ppsdriver, using a FAST_INTR, running in single user mode with no other programs, and only a tickle of data stored to the disk from the measuring program. The input event rate was 1Hz. Total samples: 86664 Standard Deviation: 250 nsec Average: 0 (dataset centered on mean) Median: -30.1 nsec Min: -763 nsec Max: 2936 nsec first column: the percentage of samples second column: +/- span around the median in nsec third column: +/- span around the mean in nsec In other words, 50 percent of the samples were inside a band of +/- 137 nsec of the mean of all the samples. 10 23 24 20 48 49 30 74 75 40 102 104 50 132 137 60 168 174 70 208 217 75 233 242 80 261 271 90 356 358 95 474 456 97.5 636 606 98 693 663 98.5 763 733 99 864 834 99.25 992 962 99.5 1215 1184 99.75 1215 1436 99.9 1642 1612 99.95 1741 1711 99.99 2040 2009 100 2966 2936 -- Poul-Henning Kamp FreeBSD coreteam member p...@freebsd.org "Real hackers run -current on their laptop." FreeBSD -- It will take a long time before progress goes too far!
--- End Message ---
FreeBSD interrupt responsetime...
Poul-Henning Kamp Sun, 2 May 1999 03:23:00 -0700 (PDT) (envelope-from phk@critter.freebsd.dk)
