UNIX for Advanced & Expert Users

I need to do some code benchmarking in unix.

On windows, it is always recommended that before you run any kind of benchmark, that you execute the command

Code:

Rundll32.exe advapi32.dll,ProcessIdleTasks

which commits idle tasks, so that hopefully this does not happen in the middle of your benchmark and distort the result. See Benchmarking on Windows XP: Home Edition and Professional for details.

Is there any analogous command for unix, or other actions that I should perform before doing a benchmark?

Please let me know if there are variations for different unices (e.g. linux, solaris, etc).

What you are asking is kind of non-sequitur for unix. Benchmarking does not measure elapsed time, use a tool like time

Code:

time my_benchmark_code parm1 parm2

It gives output like this:

Code:

real     0m0.44s
user    0m0.21s
sys     0m0.01s

real = wall time
user = time in user mode
sys = time in kernel mode

If your code is printing out elapsed times this will not work.

Thanks for the tip--I was not aware of that command. I can see that it would be very useful for all kinds of benchmarking.

Unfortunately, it is not useful for my case, however. Maybe I should have put more details in original posting: I will be benchmarking various random code from within a Java program, and Java already offers APIs for accessing wall clock, system, and user times. So I already have all that functionality--in nice wonderful platform independent form.

I was just wondering if unix had the same idle task issue (and similar solution) as windows does.

Note: I will switch to using the term idle process below, as it looks like idle task in unix refers to a real specific process. (Maybe background process or daemon or something else would be a better term?)

You say:

A non-sequitur is a conclusion that does not follow from its premises--I was not aware that my mere inquiry had either premises or a conclusion!

Maybe what you mean is that my question simply has no meaning in unix. Is that because unix has no concept of idle processes like windows? Or is it because there is no way that idle processes can suddenly come alive in the middle of your benchmark like they can in windows?

The web searching that I did on the time command said nothing about it first committing idle processes, by the way.

Correct - it makes no sense in unix. You do not "commit idle processes" in unix.
If I understand what you mean, init does that on it's own.

Question: why would you "commit" idle processes before benchmarking, when in real life whatever you are benchmarking would probably have to run when these idle process run?

Did you read that Microsoft tech link that I included in my first post?

"Commiting idle tasks" in windows means that idle (usually low priority?) processes, which have not completed yet, are suddenly given priority and run.

You want to do this before a benchmark so that no such idle task suddenly comes alive in the middle of your benchmark for whatever reason and contaminates your benchmark.

Thanks for pointing out init. Are you referring to the init process, or the init (system) command? There appears to be a distinction (Unix for Advanced Users - The Unix Boot Sequence - Starting init).

The web research that I did just now does not seem to say anything about init being able to cause idle processes to be forced to run, but maybe I missed something.

Why? Reproducibility of results.

Generally, when you publish benchmarks, you do so running no other programs. You may also try to turn off many background services/processes/whatever.

You do this so that other people running your benchmark on the same hardware are more likely to get the same result as you.

The more stuff besides your actual benchmark that can possibly be running during benchmarking, the less reproducible your results become.

The exception to these comments are when you specifically want to see how your task runs on a loaded system of some kind, but this is less common.

If unix simply lacks the functionality that windows has in this area--fine. Idle tasks suddenly running in the middle of your benchmark is usually a rare event. Its certainly more important that you do obvious simple things like make sure that no other non-idle tasks are running. But having an api for committing idle tasks like windows does certainly is icing on the cake, as it can eliminate the rare possibility.