分布式key-value系统错误污染检测

发布时间：2018-11-25 15:41

【摘要】：随着key-value存储系统的广泛使用,越来越多的研究开始关注分布式系统中的可信问题,其中一个重要的问题是,如何在系统被入侵或者管理员配置错误并运行一段时间后,检测出受污染的数据,从而可以在恢复错误数据的同时保留系统的合法更新.文中提出了一种基于key-value存储系统的错误污染检测方法,该方法允许在客户端不可信的前提下,检测客户端之间的污染扩散.文中设计了一种基于各服务器逻辑时钟的向量时钟,该时钟以用户关联操作为更新规则,记录了跨服务器的读写请求逻辑关系,用以进行错误污染跟踪;同时为了减少大规模系统中污染检测的开销,基于该向量时钟,文中进一步提出了一种在分布式系统中由操作序列构成有向无环图的污染分析方法.基于广泛使用的Voldmort key-value系统,文中实现了一个错误污染检测系统,Tracker Store.在集群测试环境下,文中对引入新的检测系统后产生的额外延迟开销进行了测试.
[Abstract]:With the widespread use of key-value storage systems, more and more researchers begin to pay attention to the trusted problems in distributed systems. One of the important problems is how to run the system after intrusion or administrator configuration error for a period of time. The contaminated data is detected so that the incorrect data can be recovered while keeping the legitimate update of the system. In this paper, a method of error pollution detection based on key-value storage system is proposed. This method allows the detection of contamination spread between clients without being trusted by the client. A vector clock based on each server logical clock is designed in this paper. The clock takes user association operation as the update rule and records the read and write request logic relationship across servers to track the error pollution. At the same time, in order to reduce the overhead of pollution detection in large-scale systems, based on the vector clock, a pollution analysis method is proposed, which is composed of operation sequences and directed acyclic graphs in distributed systems. Based on the widely used Voldmort key-value system, an error pollution detection system (, Tracker Store.) is implemented in this paper. In the cluster testing environment, the extra delay overhead caused by the introduction of the new detection system is tested.
【作者单位】：清华大学计算机科学与技术系;清华大学信息科学与技术国家实验室;
【基金】：国家自然科学基金(60925006) 国家“八六三”高技术研究发展计划重大课题(2009AA01A403)资助~~
【分类号】：TP309;TP333

【共引文献】