Font Size: a A A

Partition Of Task Type Based On Resources And Real-time Requirements And The Research Of Load Balancing On It

Posted on:2012-08-08Degree:MasterType:Thesis
Country:ChinaCandidate:X D RenFull Text:PDF
GTID:2178330335950758Subject:Computer system architecture
Abstract/Summary:PDF Full Text Request
In recent years, with the rapid development of information technology, the distributed systems technology has become a research hotspot in various important scientific researches. The tasks executed in parallel on a distributed system could running quickly and efficiently, get the result that meet user needs, In addition to the scale and rely on the system hardware and software environment outside, the key lies in the scheduling and management of workload. Load balancing is an important technique in distributed systems, using load-balancing technology able to balance the allocation of tasks to different server nodes, thereby improving overall system capacity and response speed.For decades, dynamic load balancing scheduling adjust the division of tasks according to the system load conditions. However, each scheduling algorithm has many constraints and application conditions. They only apply to certain situations and specific environment, although researchers have made a number of load balancing strategies and algorithms, however, it is difficult to achieve in improving the system utilization, based on load balancing, experimental environment and for different objectives, the algorithm is not strong general. Load balancing algorithm for task allocation has been a research hotspot in the cluster system. In the Load balancing algorithm, researchers have proposed many effective strategies. However, most existing algorithms have some drawbacks.Weighted least connection schedul algorithm is the classical algorithm in the Load balancing algorithm, which has been widely used. There are still inadequacies in the following:1. Because of different types of tasks user requests, the use of machine resources is also different. Therefore, with the current number of connections to signify the load conditions of each server node is not accurate.2. To each server is assigned a fixed weigh according to the actual hardware configuration of each server case, the weights are set manually by the administrator, to rely on experience as a rough estimate, not accurately reflect real-time processing of server nodes. Also, not dynamically modify the weights in the process of scheduling.For these two issues, here is no better way to resolved. To address these issues and to better improve the processing power of distributed systems, in this paper, weighted least connection scheduling algorithm is analyzed, through analysis we have come, different task types rate of server node resource utilization and the server's current load condition should be taken into account, and then task scheduling, in order to improve overall system performance. Based on this idea, the analysis of different types of the task of user requests different effects on the machine load, from this point on the weighted least connection algorithm is improved, proposed task-based load balancing algorithm. New algorithm can be simply summarized as follows:First, the client task is divided into three types:CPU consuming, that a large number of computing tasks based instruction; I/O consuming, that is, a lot of reading and writing instruction; Real-time type, that is demanding on the response time task. Then each type of task to assign a priority queue, for various types of tasks according to their priority rules into the team, waiting for scheduling. When the mission arrives, we cannot predict the type of tasks, will not be able to predict what kind of resources the task needed. Therefore, when task scheduling, judge the degree of each node busy from the perspective of resource utilization, when select the load parameters, CPU usage, I/O wait time should be into account, in order to better complete the scheduling. So we have to real-time monitoring The status of tasks connections, CPU utilization and the average I/O wait time of each server node, to each server node, respectively, from small to big according to the number of connections, CPU occupancy rate low to high, I/O average waiting time from short to long queues are arranged in three. Finally, distribute the appropriate type of server in the server queue to the appropriate types of tasks. Because of real-time based tasks are very strict requirements on time, therefore, we should first type of task queue for real-time task allocation in the machine, be the first to respond in real time based tasks. For I/O consuming task, customers need to wait a long time, in order to reduce customer wait time, compared with the CPU-consuming tasks, we should be for the I/O consuming task in task queue allocation machine. Finally, distribute of machine for the CPU consuming tasks. The priority within of the I/O consuming task queue and CPU consuming task queue, by I/O wait time from big to small, CPU occupancy rate from low to high ranking, the priority within of real-time tasks queue according to the urgency of the task and the time these two parameters calculated.After the presentation of the load balancing algorithm which is based on distinguishing task type, we theoretically proved that the new algorithm as a load balancing algorithm. By comparison experimental data of new algorithm and weighted least connection scheduling algorithm, we can make a conclusion that the new algorithm better than the weighted least connection algorithm for full use of node resources and improve response time.In the analysis of the current cluster system for data retrieval requirements, this paper designed a data retrieval cluster system architecture model, also, the proposed type task-based load balancing algorithm is applied to the system's task scheduling.
Keywords/Search Tags:Cluster system, dynamic load balancing, weighted least connection scheduling, partition of task type
PDF Full Text Request
Related items