Dynamic Scheduling and Load Balancing for Monte Carlo based Radiotherapy Simulations

Project Details

Project Lead
Akash Pargat 
Project Manager
Akash Pargat 
Project Members
Ravi Vadapalli  
Institution
Texas Tech University, High Performance Computing Center  
Discipline
Computer Science (401) 
Subdiscipline
51.1399 Medical Basic Sciences, Other 

Abstract

The Monte Carlo dose transport computations are intractable for small cluster computing environments. We have simulated prostate cancer plans using 25X10^6 200-MeV protons. In the previous work using the medical grid for these computations, we reported significant loss in grid speed-up and grid efficiency. We believe that dynamic scheduling and load balancing strategies could improve simulation efficiencies and throughput in this type of data parallel applications. Our initial numerical experiment using cluster computing are promising. Additional calculations are in progress. Implementation of cluster-level dynamics scheduling and load balancing could offer overall improvements to the grid-scale speed-up and efficiency.

Intellectual Merit

N/A

Broader Impacts

By this experiment we can say that, by doing the load balancing and dynamic scheduling one can reduce the estimated simulation run time. Thus can help in efficiently utilizing the resources a firm have.

Scale of Use

I want to run a simulations on entire systems and for each I'll need about 10 days to do that. I will then compare the result and after comparing