[Gems-users] Simics unstalled with processor binding

["James Poe" <gemsmaillist@xxxxxxxxx>]

Hi all,

I am trying to run simulations using (an unmodified) LogTM protocol, but a large number (~20%) of them keeping dieing (Simics becomes Shaky) and the log output shows:

Fatal Error: in fn MemoryTransactionResult SimicsProcessor::makeRequest(memory_transaction_t*) in simics/SimicsProcessor.C:354: Error: Simics get unstalled without ruby asking?

I've read the previous posts on this issue, and looking at the trace files I do believe the cause is a thread that is unresponsive.  I'm using processor binding, however, so I'm not sure what might still be the cause of the problem.  What's weird is that I'm running most of the Splash benchmarks with various processor numbers, and the issue will occur somewhat randomly for different configurations - meaning that it might crash with 4 and 16 processor versions of cholesky, but not 8, etc.  For some of the benchmarks if I try to slightly modify the image (
e.g. run an 'ls' command before invoking the benchmark) it will run correctly (I assume it just happens to come together in a manner that it doesn't generate the problem).  The problem with this is that it doesn't always work, and takes a long time to repeatedly run many simulations just to get one to work.  Also, certain benchmarks (such as raytrace and water-nsquared) always fail.  

Do others experience this issue this often with processor binding enabled?  Are there any other methods that I might use to try and stop this from happening?

Thanks as always,

James P