The Study Of Packet Scheduling And Flow Control For Improving Switch Functions

With the rapid development of networks,network applications become more popular among users.Due to the more complex and fickle user requirements,the disadvantages of switch functions are prominent.To improve the quality of network service,switches provide differential services to allow unfair competition of network resources among flows.Data flows with strict performance limitation such as latency-sensitive flows can be processed first.Unfortunately,performance requirements are multipledimensional.For example,the packet with network control messages should be processed immediately to keep robustness of the network,the packet with a deadline should be processed before it exceeds the time limit otherwise the bandwidth used will be wasted.What's more,packets wait in switches before being processed or sent out.Transmission latency can increase significantly if there are too many waiting packets,which decreases the benefit of the network.Try to meet all the user requirements and maximize network benefit,this paper analyses packet scheduling structures and queue length control algorithms of switches,and improves these functions according to their disadvantages.The main contents of this paper include two parts:· The first part,since modern scheduling structures only make decisions on single dimension values,this paper proposes a new scheduling scheme Proteus,which supports analyzing multiple-dimensional information to decide scheduling order.This design fully considers different characteristics of various data packets,try to give a proper scheduling order to meet user requirement of multiple dimensions.This work starts at a specific example,then introduces the structure and scheduling algorithms of switches.Proteus applies min-heap to maintain quantitative dimensional information and provides the whole perspective for switches to make decisions.In the meanwhile,Proteus is compatible with modern single dimension scheduling methods.According to simulation and testbed experiments,Proteus does not bring too many extra burdens of switches under single dimension scheduling,and Proteus can decrease executing time by around 80% under multiple dimensions scheduling.· The second part,focus on limiting queue length increasing with flow control under lossless network,this paper improves PFC mechanism which keeps network lossless to decrease tail latency caused by long queue length.This work involves prediction,proposes a P-PFC algorithm that predicts queue length build-up,and triggers pause message to limit data transmission before it happens.This work provides theoretical analysis to prove the feasibility of P-PFC,and gives the detailed functions of prediction and flow control.Simulation experiments show that P-PFC decreases maximum queue length and tail latency significantly,which then improves the transmission efficiency of the whole network.