[openib-general] questions and a comment on the perftest package

[Michael S. Tsirkin]

Quoting Or Gerlitz <ogerlitz at voltaire.com>:
> Following the above suggestion would enable to print the avg,min,max,median,std
> statistics where now there is this delta approximation which i find less informative.

I'm not very interested in adding the average/std statistics - our measurements
have too high a deviation for these to be useful.  There's the -H option
so you can always feed the results to your favorite statistical package.

In my experience, people ignore the std deviation and immediately zone in on the average
as the main result - so I get reports like "my latency went up by
2micro, help!" when there was one freak huge value that throws the avg off,
and the std is on the order of 10 micro.

Median does not have this problem - you get a representative value no matter
what. And in practice in my measurements with default values, it is stable to
within 0.1 micro.

-- 
MST