Study Of Packet Scheduling Algorithm In Hybrid Optical-Electronic Switching For Large-Scale Data Center

With the rapid growth of the cloud and big data applications,the traffic in datacenter far exceeds the traffic between datacenter and external users,which has brought unprecedented challenges to the network infrastructure.In this situation,optical switching becomes a feasible way to meet requirements of the large bandwidth and low latency.Compared with electrical switching,optical switching not only provides a large bandwidth but also reduce the link delay.However,it is too long for optical switches to establish or dismantle links.Electrical buffer and packet scheduling are used to make up for this deficiency.Based on the two technologies,this paper provides a new solution to build a scalable datacenter with high performance.Firstly,the article proposes a new way of delay-differentiated scheduling in optical packet switches.Secondly,we design new packet scheduling algorithms in multistage switching network and analyze the performance of them.In order to meet different demands of cloud application services,data centers need to provide different Qo S guarantee for each application.For example,a real-time video streaming usually requires strict time delay and adequate bandwidth guarantee,however,applications such as mail service is relatively delay tolerant.To support these different services,data switching among different modules must be more diverse.In this paper,we focus on the delay-differentiated packet scheduling in crossbar input queued switches.We propose two heuristic algorithms which can guarantee 100% throughput and satisfy the differentiated delay constraints.The core idea is that we decompose the traffic matrix into a series of weighted permutation matrixes and decide the packet transmission order on account of the delay constraint matrix.Our simulation validated the effectiveness of the two algorithms.Due to the diversity of applications,datacenter must deal with unbalanced traffic demands.This paper considers unbalanced traffic distribution in a three-stage Clos switching network to ensure 100% throughput and minimize traffic delay by balancing bandwidth utilization.We propose two heuristic algorithms to decompose the traffic matrix and make the most of empty time slots in each configuration by taking a multiple hops strategy.Our simulation results validated the better performance of the multiple hops strategy by comparing two proposed algorithms.