A Parallel Method for Filesystem Consistency on Supercomputing Clusters
SESSION: Research Poster Reception
EVENT TYPE: Poster
TIME: 5:15PM - 7:00PM
AUTHOR(S):James C. Ianni
ABSTRACT: Modern day supercomputer cluster systems can be hard to maintain, especially the various types and vast number of files that can reside on the cluster local filesystems. Missing files or old library versions with the wrong permissions set or wrong links can wreak havoc on the execution of parallel software. The ability to identify such missing or inconsistent files amongst the unique files that may also reside on the system is importance. There are some commercial software packages that can alleviate some problems. These packages cannot identify unique and unimportant files, operate in an efficient and parallel manner, require prebuilt, centralized databases or interfere with executing parallel user jobs on the supercomputer. We will demonstrate a very different parallel file system verification method that has been successfully utilized at the Army Research Laboratory’s DoD Supercomputing Resource Center for the past three years and does not exhibit the before mentioned shortcomings.