Font Size: a A A

Ip Network Transmission Link Congestion Control And Cluster Server-side Performance

Posted on:2008-09-16Degree:MasterType:Thesis
Country:ChinaCandidate:X W XieFull Text:PDF
GTID:2208360215486698Subject:Computer software and theory
Abstract/Summary:PDF Full Text Request
With the development of Internet, congestion happens not only in transmission link. As the QoS of Internet is end to end,on the whole route from user end to server end,nearly every position may becomes a bottleneck,including network,server end,even user end itself. The question of bottleneck in server end and in transmission link are our major ones. This article analyses this two questions and gives solution to them.To solve the question of bottleneck in server end, this article gives a NASACS(New Auto Self-Allocating Cluster System).It designs a new TCP connection mechanism and a new scheduling frame which simplity the function of dispatcher and reduce the burden of dispatcher. Moreover, we increase network card in server to specialize the transmission link in the internal part of cluster system,to promote transmission performance and to ameliorate NASACS,and adopt capacity control mechanism on the basis of operation sequence model in the back server,and adopt task admission control on the basis of prority and allocation control in the front server. In the NASACS,back server obtains connection request in dispatcher according to its own idle degree and capacity control mechanism. The work's way of back server makes sure the load's reasonable allocation, so it needn't the load balancing any more.As to the question of bottleneck in transmission link, this article studies AQM algorithm. On the basis of RED algorithm,we make simple andeffective improvement-design a power function,with the aim ofreplacing the linear function which calculate drop proportion in the original RED algorithm,and others remain unchanged. We call the new algorithm PF-RED algorithm. When the load is light, the drop proportion in PF-RED algorithm is lower than in RED; When the load is heavy,the case is reverse. With the growth of load,drop proportion's speed becomes more and more faster instead of remain a stable growth speed .The simulation experiments on the basis of OPNET testify that the control mechanism makes the throughput of bottleneck link improve,and low the sensitivity degree of simulation and queue length to parameter max_p greatly.
Keywords/Search Tags:Service Quality, end-to-end performance bottleneck, Cluster Server, Load balancing, Congestion Control
PDF Full Text Request
Related items