[parsec-users] Benchmarks which do significant amount of IO

Christian Bienia cbienia at CS.Princeton.EDU
Mon Aug 17 16:47:10 EDT 2009


Hi,

 

Some PARSEC benchmarks do some disk I/O, however it's never much or very
interesting I/O patterns because PARSEC is predominantly a CPU benchmark
suite. A lot of disk I/O would limit CPU utilization and thus the maximum
achievable speedup due to Amdahl's Law.

 

Benchmarks that perform some I/O in their ROI are bodytrack, facesim,
ferret, vips and x264. Dedup also performs disk I/O but uses preloading by
default to move the disk I/O out of the ROI (this can be disabled).
Typically it's synchronous I/O, but some benchmarks use other strategies to
hide some of the I/O latency such as running the I/O in a separate thread or
pipeline stage to overlap it with execution.

 

Best,

Chris

 

 

From: parsec-users-bounces at lists.cs.princeton.edu
[mailto:parsec-users-bounces at lists.cs.princeton.edu] On Behalf Of Vishal
Ahuja
Sent: Monday, August 17, 2009 2:07 PM
To: PARSEC Users
Subject: Re: [parsec-users] Benchmarks which do significant amount of IO

 

Also, it would be great to know if the io is done in synchronously or
asynchronously. Thanks. 

2009/8/17 Kotecha, Grishma <grishma.kotecha at intel.com>

Hi 

 

I have a question related to this:

 

When I run Bodytrack and observe CPU utilization, I see it does not get
fully utilized (it gives performance somewhere near 40-45%).

 

Is lot of memory read/write a reason for this?

 

 

Grishma 

  _____  

From: parsec-users-bounces at lists.cs.princeton.edu
[mailto:parsec-users-bounces at lists.cs.princeton.edu] On Behalf Of Vishal
Ahuja
Sent: Monday, August 17, 2009 10:00 AM
To: parsec-users at lists.cs.princeton.edu
Subject: [parsec-users] Benchmarks which do significant amount of IO

 

Hi,

       Can someone please point out the benchmark programs which do a
significant amount of io (disk read/write).

Thanks.


_______________________________________________
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/20090817/ee32b2be/attachment-0001.html>


More information about the parsec-users mailing list