面向会话的负载均衡简化算法  

Session-Oriented Fast Load Balancing Algorithm

作　　者：龚俭[1] 陆晟[1] 芮苏英[1] 

机构地区：[1]东南大学计算机科学与工程系,江苏南京210096

出　　处：《小型微型计算机系统》2005年第10期1693-1698,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金重点课题(90104031)资助国家"九七三"课题(2003CB314803)资助.

摘　　要：负载均衡算法被广泛应用于并行处理、服务集群等环境中．一些基于网络报文内容相关性的应用,例如IDS和IPv6的 Anycast服务等要求在对报文进行负载均衡分配时要保持网络会话的相关性,即相关的报文要分配到同一个处理节点,否则其 语义不能得到正确处理．传统的负载均衡算法对于这类服务需要在会话的上下文信息规模和会话完整度之间权衡,对于会话数 量很大的情况通常开销也很大．基于位熵的概念,本文提出了一种可满足会话完整性的负载均衡简化算法-城分类算法．该算 法不需要各处理机之间内部通信协调工作,也不需要在调度节点保持会话的上下文,在满足报文或会话相关性要求的同时,仍 能保持较好的宏观平衡度和微观平衡度．Load balancing is widely used in parallel computing and cluster computing environments. Some packet content oasea applications, e.g. IDS and Anycast, require network session completeness while balancing the traffic load. That is, packets belonging to one session should go to the same processing node, otherwise the semantic of the session will be violated. Traditional load balancing algorithms have to do a tradeoff between session context size and session completeness for such applications, which will put high demands on resources when the number of session is large. Based on the concept of bit entropy, a load balancing algorithm named Dimension-based Classification Algorithm is suggested in this paper. With this novel method, neither the internal communication among procession nodes, nor the storage of session contexts are required. This algorithm has a fairly good load balancing performance in both macroscopical and microscopical scopes, while keeping the completeness of sessions or segmented packets.

关 键 词：负载均衡 报文分类 位熵 入侵检测 高速网 

分 类 号：TP393.3[自动化与计算机技术—计算机应用技术]