Inquiry about Parallel Efficiency on Cluster with Aspect Code

ASPECT
1

Dear Aspect Developers,

I hope this message finds you well. I am a beginner in Aspect and have recently completed initial studies on the software. I have developed a code simulating slab subduction, which runs correctly on a local machine. However, when attempting to scale the computation on a cluster with 110 cores, the performance improvement is only about twice as fast compared to running it on an Intel i9 processor with 20 cores.

Given the significant resources allocated, this efficiency seems lower than expected. Could you kindly advise on potential causes for this scalability limitation? For your reference, I have attached:

  1. Details of the model’s degrees of freedom.
    Number of active cells: 7,168 (on 3 levels)
    Number of degrees of freedom: 335,800 (59,666+7,637+29,833+29,833+29,833+29,833+29,833+29,833+29,833+29,833+29,833)
  2. The relevant code snippet.
    test-case-4.prm (8.8 KB)

I would greatly appreciate any insights or suggestions to optimize parallel performance. Thank you for your time and support!

Best regards,
Yuan

2

Yuan,
In order to achieve parallel efficiency, you will likely need ~100,000 degrees of freedom per core. This may vary somewhat depending on the CPU, memory architecture, and interconnect on your cluster.

Max

3

Thank you very much for your answer. It really helped me a lot.