Fundamental Study on Parallel Archiving System and Adaptive Load Balancing Mechanisms for it

为了建立一个下一代所需的超大规模的Peta字节类档案库，通过结合较小的TB级的商品档案库，我们开发了高效的文件管理算法。本文首先研究了并行归档系统的可行性，阐明了可物理实现的迁移算法，并对其代价和性能进行了估算。我们设计了一个由数百到数千个元素档案库组成的并行归档系统的概念结构，并针对元素档案库之间的磁带迁移提出了基于访问频率的自适应算法。我们指出，考虑到每个元素存档器的处理能力，即使某些磁带机发生故障，也可以在元素存档器之间自动重新分配磁带并为请求提供服务。此外，还审查了元素档案库在加快数据I/O操作方面的合作。此外，利用全球环境数据库系统的文件组织和访问模式，阐明了并行归档系统在实际应用中的有效性。

To establish the method of constructing a very large scale archiver of peta byte class which is needed at the coming generation by means of combining relatively small commodity archivers of tera byte order, we develop the efficient file management algorithms. Cooperation of element archivers ranging form hundreds to thousands, file management and file migration on them are not studied yet.At first, we studied the feasibility of parallel archiving system, clarified the migration algorithms which could be realized physically and estimated the costs and performances for them. We designed a conceptual architecture of parallel archiving system consisting of hundreds to thousands of element archivers and proposed the access frequency based adaptive algorithms for the tape migration between the element archivers. We indicated that considering the processing capability of each element archiver made it possible to redistribute tapes automatically among the element archivers and to serve requests even if some tape drives fails. Further, cooperation of the element archivers to fasten data I/O operations was also examined. Moreover, using the file organization and the access pattern at the global environmental database system, we clarified the effectiveness of the parallel archiveing system for real applications.