Research On The Key Technology In Distributed Computing System

With the development of computer technology and the popularity of network, we desire more strongly for lager scale computation, storage and transmission.But, the improving hardware and software for single computer still can't meet these applications. So, to integrate multi-compter hardware to an entire system with software which complete a large scale task has become more and more urgent. The distributed system emerges in order to adapt the increasing need for large scale system and the development of the computer network and software technology. In the field of scientific research and so on, more and more in-depth and extensive application of large-scale distributed computing systems that take full advantage of the INTERNET network, to build a WAN environment distributed computing systems which component element is heterogenous.This thesis is based on a large scale gene analysis project-"DNA Chip and Pig&Fowl Gene Analysis", which is a large scale computing project consumed lots of computing resource with long time running. And, the computing load and the memory load are different among different computing sub-tasks, at the same time, the project has different computing requirement for these sub-tasks. Because the traditional small and middle scale cluster hasn't meet the project's ocean computing and the cost of special large scale cluster is expensive, we adopt the idea which is distributed computing system. In the distributed computing system applications transparent memory usage,load-balancing technology is the focus of previous research literature, However, the lack of different sub-task corresponding management in these literature, Therefore, this issues for the implementation of projects in computer applications, the multi-priority transparent memory usage, the multi-priority load balance(sharing) and the multi-queue node scheduling policy are three key technology further research by this thesis,which protects the node performance so that the effect to the local performance is limited to 5%, at the same time, meets the different sub-tasks qulity of service requirement with the CPU and memory resource, and balances the proportion among different property sub-tasks in the nodes.The multi-priority transparent memory usage algorithm proposed by this thesis not only assures the project application transparent to local node,but also the memory is allocated among different priority sub-tasks reasonably, which meet the relative qulity of service requirement among sub-tasks;The multi-queue based node scheduling policy which adopts multi queues to provide the non-preemptive service for the transparence of computing resource At the same time, the algorithm also provides the preemptive service based on priority for the different priority sub-tasks, which meet the relative requirement of service qulity comprehensively with the transparent memory usage;The multi-priority load balancing algorithm, which makes the priority of tasks and the usage of memory and CPU as the evaluation factor to improve the memory load balancing underlying the computing load balancing. Further, the algorithm also optimizes the allocation of tasks with different priority.