Parallelization of Radiation Therapy Treatment Planning(RTTP) : A Case Study

Radiation therapy uses ionizing radiation to treat cancerous tumors. This paper reports our experiences with the parallelization of a real-world 3-D radiation therapy treatment planning (RTTP) system on a wide range of platforms, including SMP servers, Cray J916 vector machines, and clusters of SMPs. The RTTP system is a meta-problem, comprising two major loosely-coupled components: dose calculation and genetic optimization. To accelerate the planning process, we integrated the two components and parallelized the system on a wide range of platforms using vendors’ native optimization tools and Stanford SUIF parallelizing compiler. For comparison, we also manually parallelized the system using multithreading, messagepassing, and distributed shared memory tools. The experimental results showed that none of the automatic parallelization tools were capable of handling the real-world application, although plenty of parallelism exists in the codes. The difficulty is due to RTTP’s dynamic and irregular computational behaviors.