Design of load balancing algorithm for CPU and GPU collaborative computing based on C4.5 decision tree

Abstract

In recent years, hybrid heterogeneous computing systems based on CPU and GPU have gradually become one of the hot research directions in academia. In practical applications, many hybrid heterogeneous computer systems based on CPU and GPU show good performance. However, due to various reasons, heterogeneous computing still faces many problems. The most prominent problem is the difficulty of program development, especially when it is extended to the cluster scale level. These problems are mainly reflected in scalability, load balancing, and adaptability. Therefore, this paper combines the current architecture of CPU and GPU, improves the C4.5 decision tree algorithm, and proposes a highly efficient dynamic weighted load balancing algorithm IRRBODT suitable for collaborative computing of CPU and GPU, so that the CPU and GPU can collaborate efficiently Computing to improve parallel computing efficiency.