摘要
覆盖网协同缓存(overlay cooperative caching,简称OCC)聚集客户节点的资源来提供可扩展、经济有效的缓存服务.在典型的OCC系统中,节点的异构性和工作负载的不对称,容易造成节点资源使用的不平衡,形成一些负载过重的"热点"节点.但对于这一问题,已有的OCC系统缺乏有效的负载平衡机制.针对多媒体内容分发服务,设计了一种无热点的OCC策略——HFOCC(hotspots free overlay cooperative caching).通过将"热点"对象复制到低负载节点,分散服务请求,达到消除热点的目的.为了提高缓存空间的利用率,HFOCC将一个节点的缓存空间动态地划分为home cache和replica cache两部分,并实施统一的缓存管理策略;基于一种"软"副本生命期控制机制,当工作负载发生变化时,冗余副本被及时删除,系统表现出了良好的自适应性.实验证明,HFOCC有效地提高了系统吞吐率和资源利用率.
Overlay Web caching (OCC) exploits resources of peers to provide scalable and cost-effective web caching service. In a typical OCC system, which is often characterized by highly heterogeneous node capacities and skewed query distributions, the resources of each node may be utilized in an unbalanced manner, i.e., some nodes are overloaded and become "hotspots". Unfortunately, there are no effective load balancing mechanisms in existing OCC systems to relief the "hotspots". This paper proposes a hotspots-free OCC scheme called HFOCC for multimedia content delivery service. Through replicating "hot" objects adaptively to lightly loaded nodes, loads are distributed more evenly across the whole network. Consequently, the hotspots are relieved. In order to utilize cache resource more effectively, HFOCC splits a node's cache space dynamically into two parts, namely the home cache and the replica cache, and manages them by a uniform policy. With a "soft" lifetime control mechanism, the redundant object replicas are deleted adaptively, and the system performs well under dynamically changing workloads. Experimental results show that HFOCC improves resource utilization and system throughput markedly.
出处
《软件学报》
EI
CSCD
北大核心
2008年第3期744-754,共11页
Journal of Software
基金
Supported by the National Natural Science Foundation of China under Grant Nos.60573106, 60402027, 60573131 (国家自然科学基金)
the Natural Science Foundation of Jiangsu Province of China under Grant No.BK2005411 (江苏省自然科学基金)
the National Basic Research Program of China under Grant No.2002CB312002 (国家重点基础研究发展计划(973)
关键词
覆盖网协同缓存
复制
负载平衡
多媒体内容分发
副本生命期管理
overlay cooperative caching
replication
load balancing
multimedia content delivery
replica lifetime management