Efficient Load Balancing Technique for Parallel Ray Tracing Using a Reservoir
Authors: Joy Alinda Madamba and Francis Joseph Seriña
Abstract
Imitating the real world in emulations, such as the effect of light, is an important research area in software applications. However, a computing machine cannot precisely imitate all the aspects of light in short processing periods. Ray tracing is a technique that renders and accurately depicts light with a high time delay. Parallelizing the computers that run the rendering is not enough to reduce its running time. To further increase computing efficiency and improve ray tracing’s applicability for mainstream purposes, load balancing must be performed. In this study, a proposed alternative load balancing technique that uses a reservoir was implemented to tackle the issue of using heterogeneous computers in the network and to minimize the communication overhead introduced by parallel applications. Results show that when using a reservoir, high speed-up is achieved at 80% initial task distribution while high efficiency is achieved between 20–75% initial task distribution. With speedup as the priority, reservoir achieves at most 87% better speed-up than the scatter decomposition and 36% than the hybrid technique. It also has at most 28% better efficiency than the scatter decomposition. Finally, the reservoir algorithm achieved at most 25% better achieved speed-up (ASU) than the scatter decomposition and 10% than the hybrid.