A Hybrid Swarm Intelligence Algorithm for Compute Cluster Selection Using Bee Colony Optimization with Random Sampling

In this paper a new hybrid swarm intelligence algorithm is proposed that blends traditional swarm intelligence (SI) framework with random sampling to have an improved search through large amounts of compute cluster data (associated with computing resource allocation and utilization) for optimum cluster selection. The proposed method is called Bee Colony Optimization with Random Sampling (BCORS) and its parallel variant is called Parallel BCORS (PBCORS). Both BCORS and PBCORS are tested using High Performance Computer (HPC) cluster specification data, and their convergence speeds in finding the optimum cluster to select are compared with alternative methods. Experimental results show that BCORS and PBCORS can sort cluster data and provide optimum computing resource allocation and selection faster than current alternatives and with less memory usage. Specifically, an average speed improvement of 68% and an average reduced memory usage of 120% are observed in comparison to parallel random sampling. The scalability of PBCORS to large amounts of data is also demonstrated by using test files with up to 70,000 entries.