Based on variable sized chunking, this paper proposes a content aware chunking scheme, called CAC, that does not assume fully random file contents, but tonsiders the characteristics of the file types. CAC uses a candidate anchor histogram and the file-type specific knowledge to refine how anchors are determined when performing de- duplication of file data and enforces the selected average chunk size. CAC yields more chunks being found which in turn produces smaller average chtmks and a better reduction in data. We present a detailed evaluation of CAC and the experimental results show that this scheme can improve the compression ratio chunking for file types whose bytes are not randomly distributed （from 11.3% to 16.7% according to different datasets）, and improve the write throughput on average by 9.7%.
传统的基于P2P的实时流传输系统中的成员管理机制有的依赖于全局成员信息,缺乏可伸缩性;有的采用随机的节点选择策略,很难获得优化的全局性能.本文提出SEBMM(Synchronization Estimation based Membership Management,基于同步评估的成员管理)机制,将对等节点的选择与网络性能和节点之间的数据同步性相关联,同时建立了一套全分布式动态节点推荐机制,使系统对网络的动态改变具有很大的弹性,并潜在地帮助系统收敛于合理的全局性能.系统模型的仿真显示了系统的正确性和有效性.
故障检测器是构建可靠的iSCSI存储系统所必需的基础组件。本文实现了一种iSCSI系统中自适应故障检测器iAFD(adaptive failure detector for iSCSI)。根据心跳(heartbeat)策略,设计了一种自适应心跳机制。故障检测器通过估计预期到达时间来提供一个探测时间,并动态地估算心跳消息超时时限,以适应系统状态的变化,减少故障检测服务的错误。实验表明,该方法与其它的故障检测方法相比,故障检测出错次数较少,检测时间较短,并能够适应高可靠计算系统状况的变化,在侦测的实时性和正确性上提供较好的平衡。
