Hi Abhishek,

Are you doing the 2D channel, or the 2.5D case?
Could it be that the size of the problem is such that using 8 cores results in very little problem per core - so kind of does not help?

Also, when in parallel you can not use the Arpack driver, only the ModifiedArnoldi will work.
Finally, in my experience, using:
    <I PROPERTY="SolverType"            VALUE="VelocityCorrectionScheme"/>
    works better in parallel than:
    <I PROPERTY="SolverType"            VALUE="CoupledLinearisedNS"/>

Hope this helps in some way,
Best,
Stan

On 10/23/2018 05:56 PM, Abhishek Kumar wrote:

Hi,

I am trying to test the linear stability solver in parallel. I did a simple test with the channel flow with a single processor and on 8 processors. I observed the following:

Time taken per 1000 steps for single processor = 2.76512s

Time taken per 1000 steps for 8 processors = 24.7811s

I expected a reduction in the computational time while running parallel. Is there any issue in the parallel implementation of the linear stability solver? Is something related with the eigenvalue solver?

I would be highly thankful for the comments on this issue.

With regards
Abhishek Kumar

---------------------------------------------------------------------------------------
Abhishek Kumar
Research Associate
Applied Mathematics Research Centre

Coventry University, Coventry CV1 5FB

The United Kingdom

Email # ac7600@coventry.ac.uk, abhishek.kir@gmail.com
---------------------------------------------------------------------------------------
_______________________________________________
Nektar-users mailing list
Nektar-users@imperial.ac.uk
https://mailman.ic.ac.uk/mailman/listinfo/nektar-users