Temporal-spatial fuzzy logic algorithm for effective dynamic load balancing in distributed 3D graphics engine

Abstract

To implement computationally-intensive tasks, typically, various levels of parallelism are used. Depending on the applications, the best possible results are produced by different parallelism approaches. Pipelining is an example of hardware-implemented parallelism; multithreading is effective for the software that supports it. In this paper, we propose a fuzzy logic algorithm for dynamic load balancing in a distributed cluster performing 3D video rendering for a simulated robotic vision tests. We present results taken from a production processing environment, rather than simulations. We have tested the algorithm within comprehensive virtual reality simulator Initiative Software Earth (ISE), a project of Computer Engineering Framework (CEF). The experimental results gathered on 50+ core Linux cluster, demonstrate that the proposed approach performs better compared to conventional methods such as static and gradient load balancing.