Jun Yeon Lee, Jae Hyun Lim

The KIPS Transactions:PartA, Vol. 8, No. 2, pp. 133-146, Jun. 2001
10.3745/KIPSTA.2001.8.2.133,   PDF Download:

Abstract

In a dynamic load distribution policies, each node gathers the current system state information before making a decision on load balancing. Load balancing policies based on this strategy can suffer from processor thrashing. In this paper, we propose a new algorithm which attempts to decrease the frequency of the processor thrashing. the algorithm is based on the integration of three components. The first, the algorithm of which determine the size of jobs be transferred. The second, negotiation protocol which obtains a mutual agreement between a sender and a receiver on the transferring job size. And the third, a symmetrically-initiated location policy. The algorithm proposed in this paper used Siman IV as simulation tool to prove the improvement of performance. I analyzed the result of simulation, and compared with related works. The mean response time shows that there are no difference with existing policy, but appear a outstanding improvement in high load. The thrashing coefficient that shows the average response time, CPU overhead and the thrashing ratio at both the receiving and sending node has been used in the analysis. A significant improvement in the average response time and the CPU overhead ratio was detected using our algorithm when an overhead occurred in the system over other algorithm. The thrashing coefficient differed in the sending node and the receiving node of the system. Using our algorithm, the thrashing coefficient at the sending node showed more improvement when there was an overhead in the system, proving to be more useful. Therefore, it can be concluded that the thrashing ratio can be reduced by properly setting the maximum and minimum value of the system's threshold queue.

Statistics
Cite this article