Re: [Gems-users] Revenge of Adventures in duplicating LogTM results


Date: Tue, 4 Mar 2008 14:11:43 -0800
From: "Jim Leek" <jrleek@xxxxxxxxx>
Subject: Re: [Gems-users] Revenge of Adventures in duplicating LogTM results


Are you referring to workloads.py? If so, you can find it in
$GEMS/gen-scripts.

  Yes, of course.  I'm sorry, I wrote this email very late last night.
 
The variables are used mainly in microbench.py to configure Ruby and in
.simics script files to setup and control simulation. Take a look at
$GEMS/microbenchmarks/transactional/deque/deque.simics to get an idea of
how they
are used.

The default values are specified in workloads.py. They are changed in
the interactive
script files that we generate using config.py and gen-scripts.py

MBENCH_ARG_STRING is used to pass argument strings to the simulated
workload.

Jayaram

ps: We do not use some(many?) of the variables in workloads.py anymore.

Unfortunately, the defaults in workloads.py are not very useful.  A lot of important variables default to 'None.'   There are a number of variables that I am sure I need the same values as LogTM used in order to duplicate the results in the Performance Pathologies paper.
For example: BANDWIDTH, or TRANSACTIONS.  I'm pretty sure I got some important variable wrong because I got the following error from Ruby when trying to run the btree microbenchmark:

end_transaction_magic_call: transaction started: 11, transaction completed: 0, transaction_limit: 10000, mem_res: 807.511719 mem_total: 977.417969 mem_ratio: 0.826168
end_transaction_magic_call: transaction started: 12, transaction completed: 0, transaction_limit: 10000, mem_res: 807.511719 mem_total: 977.417969 mem_ratio: 0.826168
simics-common: log_tm/TransactionVersionManager.C:208: bool TransactionVersionManager::addLogEntry(int, int, Vector<uint32>&): Assertion `m_logSize[thread] < (m_maxLogSize[thread] * 1024)' failed.
***  Simics getting shaky, switching to 'safe' mode.
***  Simics (main thread) received an abort signal, probably an assertion.
Starting command line. (May have skipped commands in script files.)
simics>

So, I thought it would really help with my ability to duplicate LogTM results if I could actually get the values that LogTM used for their workloads.py variables.  (Actually, this is also try for the values in default_param.h, but I think I figured those out since they are documented pretty well in the wiki.)  So, please let me know.

Thank you,
Jim

[← Prev in Thread] Current Thread [Next in Thread→]