oparallel in Xflow software, number of cpus

wodeqiche2007 · September 29, 2021, 12:12

I'm using Xflow 2020x software.
Yesterday’s calculation of more than 400,000 grid CPUs used 90%, but today’s calculation of 570,000 grid CPUs only used about 30%

I want to know how It can use more CPU during analysis and so reduce the calculation time?

LuckyTran · September 29, 2021, 21:06

If there's too much overhead then the cores sit there waiting for memory to load and don't do any work.

But also check with your job scheduler and make sure there aren't any job collisions. Collisions become exponentially more frequent the more you load a cluster and you can imagine all 569 999 cores have finished and are all waiting for that last core to finish its calculation because it is running on a node that already has something else going on.

September 29, 2021, 12:12	oparallel in Xflow software, number of cpus	#1
wodeqiche2007 New Member Dung Feng Join Date: Jul 2012 Posts: 1 Rep Power: 0	I'm using Xflow 2020x software. Yesterday’s calculation of more than 400,000 grid CPUs used 90%, but today’s calculation of 570,000 grid CPUs only used about 30% I want to know how It can use more CPU during analysis and so reduce the calculation time?

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
[snappyHexMesh] snappyHexMesh sticking point	natty_king	OpenFOAM Meshing & Mesh Conversion	11	February 20, 2024 10:12
Decomposing error simple method	DanGode	OpenFOAM Pre-Processing	1	July 27, 2021 07:58
Foam-Extend 4.0 simpleFoam motorbike parallel error?	EternalSeekerX	OpenFOAM Running, Solving & CFD	0	May 10, 2021 05:55
AMI speed performance	danny123	OpenFOAM	21	October 24, 2020 05:13
[blockMesh] --> foam fatal error:	lillo763	OpenFOAM Meshing & Mesh Conversion	0	March 5, 2014 11:27

September 29, 2021, 21:06		#2
LuckyTran Senior Member Lucky Join Date: Apr 2011 Location: Orlando, FL USA Posts: 5,747 Rep Power: 66	If there's too much overhead then the cores sit there waiting for memory to load and don't do any work. But also check with your job scheduler and make sure there aren't any job collisions. Collisions become exponentially more frequent the more you load a cluster and you can imagine all 569 999 cores have finished and are all waiting for that last core to finish its calculation because it is running on a node that already has something else going on.