IBM Hardware: Test speed of an execution core reliably. | Unix Linux Forums | High Performance Computing

  Go Back    


High Performance Computing Message Passing Interface (MPI) programming and tuning, MPI library installation and management, parallel administration tools, cluster monitoring, cluster optimization, and more HPC topics.

IBM Hardware: Test speed of an execution core reliably.

High Performance Computing


Closed Thread    
 
Thread Tools Search this Thread Display Modes
    #1  
Old 08-22-2013
Devyn Devyn is offline
Registered User
 
Join Date: Feb 2005
Last Activity: 27 May 2014, 2:04 PM EDT
Posts: 34
Thanks: 0
Thanked 1 Time in 1 Post
IBM Hardware: Test speed of an execution core reliably.

Hey Folks,

Doing simple floating point or integer arithmetic is limited since if another execution core is not busy, the system will (presumably?) assign CPU resources to where they are needed so I could be getting the performance of 2 or more cores theoretically?

Any good reliable way to benchmark this on IBM Hardware?

For example, some older Power systems come rated with higher MHz rates but I want to see that advantage, if applicable, in a simple test. Anyway to do this through the CLI without having to install anything new or big?

Cheers,
DH
Sponsored Links
    #2  
Old 09-12-2013
figaro figaro is offline
Registered User
 
Join Date: Jan 2007
Last Activity: 24 October 2014, 1:59 AM EDT
Posts: 824
Thanks: 92
Thanked 25 Times in 22 Posts
Assuming you are writing your own software, no, you have to actually program the so called thread allocation yourself.
Making an even bigger assumption that you are using C++, it is probably easiest to work with a library such as Boost or Intel TBB.
You can use the command top to see what your processes are doing.
Sponsored Links
    #3  
Old 09-12-2013
Corona688 Corona688 is offline Forum Staff  
Mead Rotor
 
Join Date: Aug 2005
Last Activity: 24 October 2014, 4:38 PM EDT
Location: Saskatchewan
Posts: 19,683
Thanks: 823
Thanked 3,352 Times in 3,139 Posts
Multithreading does not work that way; two cores can't cooperate to run a single-threaded program faster. Single-threaded programs will run slower on machines with lots of slower cores; we've had some puzzled folks ask us why their new machines have worse single-threaded benchmarks than their old ones.

But more cores means you can run more threads or processes at once without sharing time; more total work can be accomplished in the aggregate; but a program has to be designed with this in mind (or multiple instances of it run) to take advantage of this capability.
Sponsored Links
Closed Thread

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

More UNIX and Linux Forum Topics You Might Find Helpful
Thread Thread Starter Forum Replies Last Post
How to test the speed of your WIFI network? freddie50 Hardware 3 10-30-2013 03:58 PM
Hardware Correction: How to change DVD write speed makh Hardware 0 10-11-2011 06:36 AM
IBM Hardware question dizman67 AIX 5 01-19-2008 04:22 PM
FTP Speed Problem on IBM P-Series equiped with AIX 5.2 aqeelcu@hotmail AIX 3 05-14-2007 06:30 AM



All times are GMT -4. The time now is 08:40 AM.