[NSWI131] HW 5: benchmarks running time

Petr Tůma petr.tuma at d3s.mff.cuni.cz
Tue Mar 24 18:06:47 CET 2020


Hi,

> I'm working on the fifth homework and I have an issue with the
> SPECjbb2015 framework. I launched it on my computer. So far, it has
> been running for 15 hours and doesn't seem to finish anytime soon.
> 
> Is it normal that this benchmark takes so much time to finish?

The typical runtime for SPECjbb2015 is around 2.5 hours, however, the exact time may vary. In any case, 15 hours is way too long and not normal.

The benchmark prints progress information - if you examine the output, you should see it gradually changing the load (denoted as IR in the output) from low to high values, up to a limit that was determined at the start of the benchmark (the limit is called HBIR or HIGH_BOUND_IR). In every step, the benchmark should display a message about running with a new IR, then a few messages about settling with the new IR, and then move on to the next step.

If you do not see the benchmark progress from low to high IR values, then something is wrong.

Also, you can try adding command line arguments to reduce the iteration range:

java -Dspecjbb.controller.type=HBIR_RT -Dspecjbb.controller.rtcurve.start=0.99 -Dspecjbb.controller.rampup.steps=3 -Dspecjbb.controller.validation.skip=true -jar specjbb2015.jar -m composite

On my laptop, HBIR is around 10000, and with the command line above, the benchmark takes about 20 minutes to finish (obviously, the results are not complete then).

Hope this helps, Petr

P.S.: This message was initially discarded because you've used a different sender address, I've added it to the list (but now you will receive messages to both your addresses).



More information about the NSWI131 mailing list