Research On Load Balance Of 3D Cluster Rendering

Highly efficient management of cluster resources is needed to exert the advantages of clusters computing. On this demand the cluster job management system appeared and fast to develop. It manages and schedules software and hardware resources according to user demand to guarantee the average share of resources and improve the system utility and output.This paper is absorbed in the aim of how to improve the efficiency of 3D cluster rendering tasks. Based on the job management system LSF, it focuses research on LSF framework, basic work process, the work principle of scheduling plug-in and optimization of extended plug-in.The resources requirement of 3D rendering is essential for evaluating the performance of hosts as we analysis the techniques of 3D rendering on clusters.Additionally, the job scheduling strategies which are supported by LSF scheduling framework and the typical resources scheduling algorithms for load balance are analyzed as well.As it is well known, the static scheduling algorithms, which depend greatly on the tasks' or servers' static attributes, couldn't distribute the load evenly among a cluster of back-end servers, due to the tasks' unbalancing and dynamic environment. An algorithm called dynamic feedback of load-increment has been designed and realized according to the dynamic feedback mechanism which will modify the load information at an interval time. This scheduling requires some mechanism for monitoring and evaluating the present load on each server so that it could select the right server to undertake the incoming task.Finally, the realized resources scheduling module is added to the LSF scheduling modules associated with LSF plug-in. Moreover, it can cooperate work with other scheduling polices in the system.