Network Scheduling Schemes And Application With Bursty Traffic

With the rapid development of multimedia application in network, the traffic becomesmore and more bursty. Since the traffic is aggregated in core network, its average ratekeeps stable over time, but fluctuates in a short period of time. The simple staticscheduling schemes cause bandwidth underutilization with bursty input traffic and thedynamic scheduling schemes require high computational complexity to deal withburstiness. Thus, a new scheduling scheme is needed to cope with the bursty traffic.Therefore, deflection-enhanced quasi-static scheduling scheme is introduced toimprove network performance in this paper. It builds up bandwidth-guaranteedconnections, which is called virtual circuit (VC), between all core network devices to dealwith the average traffic. Due to the bursty traffic, sometimes traffic overflows in VCs, butsometimes bandwidth is underutilized. Therefore, deflection mechanism is introduced toserve overflow traffic with spare bandwidth of other VCs. In this way, the bandwidth ofdifferent VCs is statistically multiplexed. It is easy to show that this scheme can achievebetter performance with larger number of VCs. To justify the effectiveness of this scheme,it is introduced to two typical bursty scenarios.Firstly, deflection-enhanced quasi-static scheduling scheme is introduced to improvethe performance of core network switches. In core switches, network traffic becomesbursty with development of network application, like high definition videos. Currently,one of the widely used scheduling schemes is Birkhoff-von-Neumann (BvN) quasi-staticscheduling scheme. Altough this scheme can achieve100%throughput and boundedend-to-end delay, its performance becomes unpredictable with bursty input traffic.Deflection scheme is introduced to BvN switch and deflection-enhanced BvN (D-BvN)switch is proposed in this paper. Fluid flow analytical model and simulation model areprovided to estimate the performance of D-BvN switch. Results show that with fixedbuffer size, deflection scheme can not only significantly reduce the packet loss rate and end-to-end delay, but also introduce little software and hardware complexity. Thoghdeflection would cause packet mis-sequence at output port, the analytical model in thispaper shows that packet mis-sequence probability can be controlled to be very low.Then, deflection-enhanced quasi-static scheduling scheme is introduced to solve theproblem of bursty traffic in datacenter. Traffic is also bursty in datacenter network. It isclassified into premium traffic which needs bandwidth guarantee and free traffic whichneeds best-effort service. Bandwidth-guaranteed VCs are provided for premium usersaccording to their peak traffic rate. Free traffic is served when premium traffic rate is at itsbottom. Overflow free traffic is deflected to other VCs when premium traffic rate is at itspeak.Since premium users have high priority in network, no premium traffic is dropped ormis-sequenced. On contrary, free users may have packet loss or packet disorder. In thisway, datacenter provides differentiated service for different users. Analytical andsimulation results show that this scheme can effectively guarantee the QoS of premiumuser. Also, it can improve the bandwidth utilization to provide best effort service for morefree users.The following are the innovative points of this paper:(1)The main idea of deflection-enhanced quasi-static mechanism is to fully utilize thespare bandwidth to serve overflow traffic. Therefore, this scheme can effectively improvethe system throughput and reduce the packet delay and delay jitter.(2) Deflection-compensated mechanism is a distributed mechanism. The schedulingof each port in switches and each edge switch in datacenters is only based on the localinformation.(3) Deflection-compensated mechanism introduces little extra hardware requirement.Also, its online computation complexity is (1).