[parsec-users] Time to Completion

Jim Dempsey jim at quickthreadprogramming.com
Thu Apr 8 10:40:29 EDT 2010


ferret using input_native on 4-core Intel Q6600 takes 648 seconds for the
Pthreads configuration using 4 threads. I do not have a timing run for
serial but I would not expect any worse than ~4x the run time of the
4-thread run. Therefore, I would expect the single thread run on this system
to be less than 45 minutes. Something is seriously wrong with run times of 8
days. Although many files are involved (3,500), ferret has a relatively low
ratio of  I/O time to compute time (using single SATA hard drive). I suspect
a pthreads issue (unless your storage is >200x slower than SATA HD).
 
Jim Dempsey
 
 
 

  _____  

From: parsec-users-bounces at lists.cs.princeton.edu
[mailto:parsec-users-bounces at lists.cs.princeton.edu] On Behalf Of
Christoforos Christoforou
Sent: Thursday, April 08, 2010 8:43 AM
To: PARSEC Users
Subject: Re: [parsec-users] Time to Completion


hey Chris,
thanks for the response.
I can tell for sure that i didnt hit the deadlock on ferret. Specifically i
ran just 2 benchmarks, ferret and x264. ferret completed after about 8 days
and exited ROI and then x264 started running, but i did not let it complete
due to the fact that i had to run some other tests without parsec.
the weird thing is that by comparing just the simulation time for ferret (8
days) and the time needed to run all the benchmarks (2.5 - 3 days) i can
only assume that i am doing something wrong while simulating all the
benchmarks instead of when simulating individual programs.


2010/4/6 Christian Bienia <cbienia at cs.princeton.edu>


Christoforos,

 

It does seem something is not going well. Could you let us know where
exactly the simulation gets stuck?

 

There are a number of things that could have gone wrong:

 

.         The programs did finish (in one way or another) but Simics did not
recognize this. This could be if you simulate until a magic break point is
reached but the program crashed or something else happened. In other words,
you continue to simulate an idle system.

.         It might be that you've hit the deadlock that was reported for
ferret. It can occur at the end of the Region-of-Interest. Chris Fensch
developed a patch for that, you can download it from the PARSEC wiki.

 

There are also other things that might have gone wrong, it's impossible to
tell anything for sure without knowing the details. Please check the output
of the simulation.

 

- Chris

 

 

From: parsec-users-bounces at lists.cs.princeton.edu
[mailto:parsec-users-bounces at lists.cs.princeton.edu] On Behalf Of
Christoforos Christoforou
Sent: Tuesday, April 06, 2010 1:59 PM
To: parsec-users at lists.cs.princeton.edu
Subject: [parsec-users] Time to Completion

 

Hello all,

I am new to parsec, and i am trying to run parsec on a sparc-solaris10
machine which is simulated in simics v3.0.31

My question is the following:

I have run all benchmarks with a simlarge load using:

parsecmgmt -a run -p all -x pre -i simlarge -n 32 -c gcc-hooks

This took about 2.5 days to complete.

Now I am trying to run benchmarks individually, e.g. x264 using the exact
same commands, the simulation has been running for 2 weeks and still hasnt
completed.

Any suggestions on what i am doing ( if i am doing ) something wrong?

Thanks in advance,

--chris


_______________________________________________
parsec-users mailing list
parsec-users at lists.cs.princeton.edu
https://lists.cs.princeton.edu/mailman/listinfo/parsec-users




-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/parsec-users/attachments/20100408/785fa4e7/attachment-0001.html>


More information about the parsec-users mailing list