[Beowulf] pthreads

[Michael Di Domenico]

does anyone know of an hpl like benchmark, but not mpi based, i need
it to be pthreads based.

i'm seeing an issue where a user is running a pthreads program that
uses a large amount of memory and many threads. the code is okay, but
when it runs it's almost as if linux is confining the job to the first
numa node rather then all the numa nodes. the oddity is that running
a taskset -c on the program does correct the issue.

i suspect numad (which i'm running) is doing something wacky, but i
have yet to prove it. the user code takes a long time to run, so i
was hoping i could find something else to reproduce the issue. sadly
most all the benchmarks i have are mpi based.
_______________________________________________
Beowulf mailing list, ***@beowulf.org sponsored by Penguin Computing
To change your subscription (digest mode or unsubscrib

[Elken, Tom]

Hi Michael,

The following is HPL-like in that it is LINPACK.

Intel MKL provides this non-MPI (most likely OpenMP thread based) version of the Linpack benchmark :
https://software.intel.com/en-us/mkl-linux-developer-guide-intel-optimized-linpack-benchmark-for-linux
" Intel® Optimized LINPACK Benchmark for Linux* is a generalization of the LINPACK 1000 benchmark. It solves a dense (real*8) system of linear equations (Ax=b), measures the amount of time it takes to factor and solve the system, converts that time into a performance rate, and tests the results for accuracy. The generalization is in the number of equations (N) it can solve, which is not limited to 1000."
As you follow this link, you will see several variations of executables/run-scripts for various types of machines.

You can read about original Linpack 1000 benchmark (for running on SMP machines) and download it from:
http://www.netlib.org/utk/people/JackDongarra/faq-linpack.html#_Toc27885717 ,
but you would want to generalize it to run on more than a 1000x1000 matrix to fill up more of the memory of your machine. The Intel version mentioned above already does that.

-Tom


-----Original Message-----
From: Beowulf [mailto:beowulf-***@beowulf.org] On Behalf Of Michael Di Domenico
Sent: Thursday, August 24, 2017 5:20 AM
To: Beowulf Mailing List <***@beowulf.org>
Subject: [Beowulf] pthreads

does anyone know of an hpl like benchmark, but not mpi based, i need it to be pthreads based.

i'm seeing an issue where a user is running a pthreads program that uses a large amount of memory and many threads. the code is okay, but when it runs it's almost as if linux is confining the job to the first numa node rather then all the numa nodes. the oddity is that running a taskset -c on the program does correct the issue.

i suspect numad (which i'm running) is doing something wacky, but i have yet to prove it. the user code takes a long time to run, so i was hoping i could find something else to reproduce the issue. sadly most all the benchmarks i have are mpi based.
_______________________________________________
Beowulf mailing list, ***@beowulf.org sponsored by Penguin Computing To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
_______________________________________________
Beowulf mailing list, ***@beowulf.org sponsored by Penguin Computing
To change your subscription (digest mode or unsubscribe) visi