On Fri, May 04, 2007 at 10:25:30PM +0300, Mindaugas R. wrote:
> yamt@mwd.biglobe.ne.jp (YAMAMOTO Takashi) wrote:
> > i don't know any ready-to-use style tools.
> > when writing priocscan, i've kludged some kernel code to measure
> > min/max/avg latencies for each i/o priorities.
> What kind of code?
> 
> Guess, maybe some kind of many small sized writes benchmark may say
> something about latencies. 10MB/s more throughput is quite much, thought...

Didn't I suggest that these changes should be benchmarked with postmark?

It doesn't matter how much throughput you gain if the system actually
performs worse for real tasks because of an increase in latency.

Thor