Re: [Gems-users] Further adventures in duplicating LogTM results (more help please)


Date: Sat, 01 Mar 2008 08:06:15 -0600
From: Dan Gibson <degibson@xxxxxxxx>
Subject: Re: [Gems-users] Further adventures in duplicating LogTM results (more help please)
Jim Leek wrote:
So, I was never able to get the Solaris 10 install to work with Simics 2.2, but my partner got it running on SImics 3.0, using the serigenti install scripts included with simics.  So, we've been working with that.  We spent tonight booting the installs with 1, 4, 8, and 16 processors and making checkpoints. 

You should be aware that we have observed different paths of execution between Simics 2.2.x and Simics 3.0.x, even with the same checkpoints. If possible, you should use your bootable disk image to make 2.2.x-compatible checkpoints. Checkpoints aren't back-compatible, but installed OS images are.
As I recall LogTM used the serengeti profile with 32 processors.  However, you can't boot serengeti with that many processors, because it only allows 6 boards with 4 CPUs each.  To get 32, we had to change the serengeti_components.py file to allow more CPU boards.  Is this correct?
Serengeti machines can be configured with multiple chassis -- each chassis sports 6 cpu boards and 2 io boards. To go beyond 32, we added another chassis.

Another confusing thing to me, is that Ruby simulates a multi-core system, so shouldn't it be 32 processors on 1 board?  Or does ruby take care of that somehow?
Simics has nothing whatsoever to do with timing. Under Simics alone, CPUs that communicate across boards, chassis, etc. do so in unity time, agnostic of how long it would take on the actual machine. Ruby enforces the timings of whatever it wants to simulate on the flat timing landscape provided with simics.

One last question, I was interested in the cholesky, barnes, and radiosity microbenchmarks, but they don't seem to have any code included.  Where can I get that code?  (/microbenchmarks/transactional/cholesky/)

Those workloads are part of the SPLASH-2 workload suite, from Stanford. Look here: http://www-flash.stanford.edu/apps/SPLASH/

Thanks,
Jim
Regards,
Dan

_______________________________________________ Gems-users mailing list Gems-users@xxxxxxxxxxx https://lists.cs.wisc.edu/mailman/listinfo/gems-users Use Google to search the GEMS Users mailing list by adding "site:https://lists.cs.wisc.edu/archive/gems-users/" to your search.

-- 
http://www.cs.wisc.edu/~gibson [esc]:wq!
[← Prev in Thread] Current Thread [Next in Thread→]