Image

Performance and Efficiency Analysis

Parallel Programming Methods (Algorithms, MPI) and Linux Clustering
User avatar
Shahram
Posts: 276
Joined: Sun Feb 05, 2017 8:31 am

Performance and Efficiency Analysis

Postby Shahram » Mon Apr 10, 2017 2:58 pm

In order to have a fast running code on cluster we need to analyze efficiency of our codes. Large number of processors doesn't mean we have fast program because we should have a balance between internal and interactive communications. some times adding more processors will reduce our speed. Beside, performance and speed is very sensitive to our algorithm too. And finally topology and structure of cluster will effect performance.
In this section we can discuss different methods to analyze performance and improve it.



User avatar
Shahram
Posts: 276
Joined: Sun Feb 05, 2017 8:31 am

Re: Performance and Efficiency Analysis

Postby Shahram » Mon Apr 10, 2017 3:02 pm

Performance Analysis of MPI Collective Operations by Pjesivac-Grbovic et al.

http://icl.cs.utk.edu/projectsfiles/ftmpi/pubs/jpg_mpi_perf_analysis_cluster.pdf



User avatar
Shahram
Posts: 276
Joined: Sun Feb 05, 2017 8:31 am

Re: Performance and Efficiency Analysis

Postby Shahram » Sun Jun 18, 2017 5:40 pm

Tip: Extra interactions (for example simple "printf") decrease speed and efficiency dramatically especially wherever we are calculating force and interaction between particles. On the other hand we'll have a huge output log. Difference is very considerable. For example adding 3 lines of output for 2 million particles changed running time from 4 hours to more than 8 hours and log file from 100 MB to 200GB !
After test remove unnecessary outputs as much as possible.




Return to “Parallel Programming and Clusters”

Who is online

Users browsing this forum: No registered users and 1 guest