Research And Implementation Of Global Resource Scheduling In Heterogeneous Multi-domain Environment

As an important part of distributed resource management technique, global resource scheduling technique in heterogeneous multi-domain environment provides quality global resource scheduling for high performance computing (HPC) service and information service. Now the distributed HPC environment is connecting several heterogeneous or isomorphic clusters, which are made up of a large quantity of PCs or workstations by specific net or Internet and installing supporting software. For managing a lot of resources every cluster has a resource scheduler with which the cluster can be a relatively independent autonomy. For this kind of heterogeneous multi-domain environment can gather many clusters, the system resource is characterized with large scale, wide distribution, heterogeneous and dynamics resulting in the complicated global resource management.Resource scheduling is the core part of resource management. Before the research about resource scheduling are mostly concentrated in one cluster, such as the LCRM, LSF, PBS etc. The study of global resource scheduling in heterogeneous multi-domain platform is not enough. The systems such as Condor-G and Nimrod/G have ever done some beneficial trails about global resource scheduling. With the continuously increasing of high performance science computing and application for large-scale distributed processing, it is high time to study the efficient global resource scheduling technique in order to strengthen the whole function of system and raise the usage of resources and jobs output.This paper based on the various existing local resource scheduler has a background of large-scale distributed resource management corresponding to the socket communication and multi-thread parallel server technique and this paper is mainly about global resource scheduling in heterogeneous multi-domain environment and take this as the foundation to carry out a prototype system of global resource scheduling. The essential parts are as follows:Establish the model for global scheduling system using the improved constraint satisfaction problems model. Implementation of concentrated global scheduling module and that of distributed global scheduling module are two different ones. Each sub-module function of the concentrated global scheduling module is discussed in detail.Study the global scheduling process and global scheduling strategy. Analyze all kinds of job supporting with FIFO scheduling strategy as its foundation. Adopt the priority scheduling and backfill strategy so as to raise the usage of system resources and the job output and guarantee the interdependent job groups execution accurately.According to the global resource scheduling in heterogeneous multi-domain environment problem we design and implement a prototype system gathering all types of heterogeneous local scheduling systems. The system consists of GlobalResourceManager and GlobalSchduler. The role, which GRM plays, is to receive and analyze the request from clients and provide the global...