Research On Scheduling Strategy Of Cluster Based On Particle Swarm Optimization Algorithm

With the gradual expansion of the cluster size and increasing performance, scheduling strategy as the key technology clusters has become a research difficulty and hotspot. PSO (Particle Swarm Optimization, PSO) algorithm has an advantage of parallelism, effective global/local search balance, simple calculation, robustness, etc, the research on cluster scheduling strategy base on Particle Swarm Optimization has high academic value and practical significance.Built a loosely coupled PC cluster based on Linux + MPI in this paper, established RAID5 data redundancy relationship between network storage nodes in the cluster, to improve data transfer rate and system throughput, Realized Logical Volume Manager (LVM), improved the flexibility of cluster system disk partition management, Reduced the mean time to data lost of node/disk while improving the reconstruction rate of the system, enhanced the security and reliability of the whole cluster system.Designed and implemented the scheduling strategy based on PSO algorithm, including the design of encoding, choose the speed-location model, determine the fitness function and constraints of the cluster scheduling problem, determine the algorithm termination criterion, design and build the particle decoding scheduling scheme. According to the research of the requirement of load balancing in scheduling strategy, and the inertia weight of Particle Swarm Optimization, Propose a equilibrium factor Ti for each node as the reciprocal of the load completion rate, By controlling the difference of load completion rate of each resource within a smaller threshold to adjust the inertia weight dynamically for ensure load balancing.Obtained the best ? , v max,? ,m values through experiment, the speed and accuracy of strategy generation are improved significantly because the excellent performance of PSO algorithm, scheduling strategy that add equilibrium factor fully into account the dynamic load or cluster system, making the overall execution time (optimal value) of the cluster smaller.Tested and evaluated the final scheduling results of the cluster, the average response time and the average CPU utilization of each node, through the contrast with FCFS and Backfill, the effect of the scheduling results based on PSO algorithm is very obvious. Greatly shortened processing cycles of the cluster system, saved the time and costs of actual production process. After adding the equilibrium factor, balanced the contradiction of high efficiency of the system and short operating waiting time greatly.