scalability issue in m_Router::initp_ in MCT

[worleyph]

This issue is to document MCT PR number 41.

Even with MCT PR number 39, ACME initialization for high resolution jobs on Mira (-compset ACME_WCYCL2000 -res ne120_oRRS15) can take over 100 minutes when using @amametjanov 's 2048 node PE layout. An optimization has been identified that drops the initialization cost to 26 minutes. Performance for this routine is also improved markedly on Titan, but the original formulation of the algorithm only takes 1-2 minutes with the target PE layouts on Titan, so the optimization is not as critical on Titan.

[worleyph]

@rljacob , upon first glance this and the other MCT optimizations are not part of the CIME5.2 update. What is the timeline for these being merged into ACME?

[rljacob]

I was going to put this in after the CIME5.2 update.

[worleyph]

Thanks. This would go directly into ACME, or does MCT go in via CIME?

[worleyph]

@rljacob ,

I was going to put this in after the CIME5.2 update.

We are up to the CIME 5.3 update. Do you have a new timeline for this?

Thanks.

[rljacob]

I'll now wait for the 5.3 update to get to master, hopefully early this week. MCT update will go directly in to ACME.

[worleyph]

I believe that this has been merged into ACME from MCT.