摘要
网络链路过载或链路失效时,使用负载均衡技术可以避免网络发生拥塞。负载的分派粒度决定了负载均衡系统的均衡性能。分派粒度越细,均衡效果越理想。基于包水平粒度的负载分派可以实现理想的均衡性能,但是会造成同一TCP业务流中报文乱序;基于流水平的分派可以保证报文不乱序,但均衡效果不理想。提出了按照报文段粒度分派负载的FSLB算法。仿真实验表明,该算法可避免报文乱序并能达到较理想的均衡效果。
Load balancing is a effective technique that keeps network from congestion when link over loading or link fai-lures.The performance of balancing system was determined by granularity of load splitting.The more fine granularity the more better performance.But splitting schemes must make a tradeoff between slicing granularity and packet reordering.Splitting traffic at the granularity of packet,each path of the balancing system can obtain accurately load assignment,but can make a lot of reordering packets in the same TCP flow.Splitting traffic at the granularity of flow,the packet of the same flow arrives at the destination along the same path,which will not cause packet reordering,but each path gets the inaccurately load assignment compared with it＇s desired load sharing.This paper showed that one could split a flow into multi path without causing packet reordering,which using the time slot between consecutive packet of the same flow to chop the flow into several segment.FSLB algorithm splited the traffic at the granularity of the segment,the results show that FSLB algorithm gets the fine performance by simulation.
出处
《计算机科学》
CSCD
北大核心
2010年第12期67-69,80,共4页
Computer Science
基金
863国家重点基金项目(2008AA01Z214)
863国家重点基金项目(2007AA01Z2a1)
863国家重点基金项目(2007AA01Z212)资助
关键词
负载均衡
报文段
流切割
Load balancing
Segment
Flow chopping
