复合加权调度算法在IAAS层的稳定性优化研究

Composite Weighted Scheduling Algorithm Study on Calculation of Stability Optimization in the IAAS layer

摘要:

多机多任务的云服务系统，在优先满足收费客户的原则下，高优先级客户的数量急剧增加，即高优先级客户数量密集饱和事件的概率将大大提高，在此情况下保证系统的QoS就成为ISSA层中关键问题， FIFO算法实现较简单，但要保证QoS机制，要利用现存的分级加权算法(Hierarchical Weighted Algorithm，HWA)才能初步实现，而基于IAAS层的高优先级客户易密集特性，HWA在该特性环境下会逐渐演变为类FIFO服务模型从而降低QoS。 我们的目标是在IAAS层中，高优先级客户请求量密集的情况下，保证满足高优先级客户请求的同时还要防止高优先级客户量增加演变为类FIFO服务模型而降低QoS，从而保证系统的稳定性。因此提出了一种基于系统稳定性的复合加权调度算法，并和FIFO以及HWA进行了比较，仿真结果表明该调度算法在高优先级客户请求量密集的情况下，系统吞吐量、平均占用均可得到明显改善，而丢包率并无太大差异， QoS能得到较好的保证，即系统稳定性能得到较好保证。

Abstract:

Under the “customer who is charged takes the priority” principle, the high priority customers of a many-processor multitask cloud service system rapidly increase. That is, the possibility of the intensiveness saturated event of high priority customer will greatly improve. Under this circumstance, to ensure the Qos of the system becomes the key point the the ISSA layer. FIFO algorithm is relatively simple, but to ensure Qos system, it still need to make advantage of the existed Hierarchical Weighted Algorithm(HWA) to initially realize. While based on the high priority customer feature of intensiveness, HWA will gradually become FIFO-like service model and then decrease Qos. Our goal is, under the circumstance of the intensive request of high priority customer in the IAAS layer, to ensure the request of high priority customer and meanwhile prevent the increase of high priority customer from becoming FIFO-like service model which will decrease Qos, thus to ensure the stability of the system. Therefore a Composite Weighted Scheduling Algorithm(CWSA) which is based on system stability is put forward. Compared with FIFO and HWA, the simulation result shows that under the circumstances of intensive request of high priority, the system throughput and average occupancy of this kind of scheduling algorithm are both greatly improved. While the packet loss probability makes no big difference, Qos could be ensured better, that is the system stability could be ensured better.