1
$\begingroup$

Most disk-based system's performance heavily depends on its disk access patterns. So maximizing the sequential I/Os while avoiding the random I/Os is a typical goal in performance perspective. I am working on the same problem.

I have some difficulties in measuring the degree of randomness of Disk I/Os. I can measure the disk throughputs in MB/sec using the performance monitoring tools. However, MB/sec is a result of such and such possible causes including the disk access patterns. I want to directly measure how much random disk I/Os are occurring at a certain moment.

Is there any metric that is commonly used to represent the degree of randomness in Disk I/Os?

If there is, how can I measure those?

$\endgroup$
  • $\begingroup$ Are you asking how to physically measure these metrics on hardware? I think that would be offtopic here. $\endgroup$ – Raphael Oct 3 '16 at 10:22
1
$\begingroup$

I am not aware of any formal discrete metrics of spatial and/or temporal locality of reference. Maybe someone else can enlighten us on that.

Often I see metrics used that are relevant to the particular platform you're studying. So the metric is either 'L1 cache misses', 'branch mispredicts', 'hard drive head movement', etc. You can google how to measure those.

You can make ad-hoc models by saving the location of your read/writes and applying penalties if those locations (or 4K blocks around those locations) haven't seen I/O recently.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.