Performance Analysis And Monitoring Of FC-SAN System For Massive Digital Resource Management

Information Explosion brings a big challenge to the storage system of Massive Digital Resource Management. The networked storage technology, such as FC-SAN, provides a solution for this challenge. One of the typical application domains of Massive Digital Resource Management is the large-scale digital library. As a system infrastructure, FC-SAN and its performance has a critical impact on data access performance and service quality of digital libraries. However, for various reasons, such as complex structure etc., current research on the performance of FC-SAN is still in a preliminary stage in terms of both theoretic works and practical tools. This dissertation, from high-end users'perspective, studies the performance analysis and monitoring issues of FC-SAN system, and presents the results of this study. The main content of the dissertation is as follows:By investigating the typical components of FC-SAN, we propose an abstract FC-SAN performance analysis model and indicate its application scope. Based on each component's operation, the model abstracts each component by a mathematic description. This process that we call SANMO2 modeling method is illustrated by a real case. The validation result of the model shows that the maximum error rate is less than 28%. By using this analysis model, we can locate the bottlenecks and the change of bottlenecks for the real case when the case system is under different workload. Besides the bottleneck locating, we also discuss the bottleneck relieving and propose the concept of Secondary Bottleneck (Cluster). We point out that, for a system that contains secondary bottleneck(s), the system resources of the secondary bottleneck cluster should all be increased, so that an effective performance improvement can be achieved.Based on the above work, a new system performance monitoring method is proposed. Compared with the current in-band method and out-of-band method, this method has several advantages such as low deployment cost, less overhead to the monitored system itself, more independent/impartial and more robust. As the effectiveness of this method depends on the expandability and the precision of theanalysis model, so we discuss this dependence also. We expand the analysis model for the cases of partial random I/O, mixture of read/write requests, and array controller cache enabling, etc. And we adjust the model to accommodate the phenomenon of Round, and discuss the applicable scope of the adjusted model. Within this scope, the model precision is almost improved by 30% (i.e. the maximum error rate is almost decreased by 8%).As a complementary to the analysis method, finally, we propose a performance simulation model MMQ for the whole FC-SAN system. This model can be applied to different FC-SAN instances. And by a real case, we also illustrate the simulation process and discuss how to analyze and evaluate the simulation results.