Scalability testing and optimization

From HP-SEE Wiki

(Difference between revisions)
Jump to: navigation, search
(SIMPLE-TS-2D)
 
Line 3: Line 3:
'' Section contributed by IICT-BAS ''
'' Section contributed by IICT-BAS ''
-
The access to clusters HPSG, Debrecen, Szegen and to SuperComputer BlueGene/P make possible to provide speed-up tests up to 1024 processes. The speed-up on Debrecen is idle up-to 768 processes, for problems of 4.5 million cells. This mean that each process has to calculate 5860 cells, which is very small number of calculations per process. The results for speed-up on other HP clusters are similar.
+
The access to the clusters HPCG, Debrecen, Szegen and to SuperComputer BlueGene/P made it possible to provide speed-up tests up to 1024 processes. The speed-up on Debrecen is ideal for  up-to 768 processes, for problems of 4.5 million cells. This mean that each process has to calculate 5860 cells, which is very small number of calculations per process. The results for speed-up on other HP clusters are similar.
-
The speed-up test on SuperComputer BlueGene/P is ideal up to 384 processes and the computational time per process is significantly larger than on HPC clusters. One process on HPCG cluster (without HyperThreding) is approximately 8.5 times faster then one process on SuperComputer BlueGene/P.
+
The speed-up test on SuperComputer BlueGene/P is ideal up to 384 processes and the computational time per process is significantly larger than on HPC clusters. We also note that one process on HPCG cluster (without use of HyperThreading) is approximately 8.5 times faster then one process on SuperComputer BlueGene/P.

Latest revision as of 09:50, 15 April 2012

SIMPLE-TS-2D

Section contributed by IICT-BAS

The access to the clusters HPCG, Debrecen, Szegen and to SuperComputer BlueGene/P made it possible to provide speed-up tests up to 1024 processes. The speed-up on Debrecen is ideal for up-to 768 processes, for problems of 4.5 million cells. This mean that each process has to calculate 5860 cells, which is very small number of calculations per process. The results for speed-up on other HP clusters are similar.

The speed-up test on SuperComputer BlueGene/P is ideal up to 384 processes and the computational time per process is significantly larger than on HPC clusters. We also note that one process on HPCG cluster (without use of HyperThreading) is approximately 8.5 times faster then one process on SuperComputer BlueGene/P.