AIM Multiuser Benchmark - Suite VII

Martin Pohlack mp26 at os.inf.tu-dresden.de
Thu Apr 12 15:01:15 CEST 2007


Hi Eduard,

> As a part of my research I have to run macrobenchmarks on L4Linux and 
> native Linux. I saw few papers related to performance testing of L4Linux, 
> where this benchmark was used. 
> The problem is that I'm not able to find (google) working version. 
> Could somebody provide me tested (used,working) copy of this benchmark?
> Or are there any similar test that could be used instead of aim7?

What exactly does not work for you?

I'm currently using AIM7 for benchmarking Linux and L4Linux und had to
patch only one line (Patch by Adam, attached).  A floating point
comparison with 0.0 does not work, which results in an endless loop.

I also changed the whole time measurement internals to gettimeofday()
and also use some performance counters etc.  AIM7 originally used
times() for measuring the single microbenchmarks.  Times() usually has a
resolution of 1-10 ms, so many microbenchmarks take only zero, one or
two milliseconds as their iteration count is meant for machines ten
years old.

Using gettimeofday(), I get a much better resolution.  However,
gettimeofday() does not give you the user / system time for the process
and or its children.  If you like I can send you my current patch.

I originally got the AIM 7 sources from sourceforge.

If I read the license correctly, you can not say you used the
AIM-Benchmark if you modify AIM's source code in any way.

Cheers,
Martin
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: aim7.patch
URL: <http://os.inf.tu-dresden.de/pipermail/l4-hackers/attachments/20070412/6e52130b/attachment.ksh>


More information about the l4-hackers mailing list