File-System Workload Scientific on a .....
Multiprocessor Kotz, Nils Nieuwejaar · NASA Technical Reports Server (NASA) · 1995
any scientific applications have intense computationalan_i UO requirements. Although multiprocessors havepermitted astounding increases in computational per-formance, the formidable UO needs of these applica-tions cannot be met by current multiprocessors andtheir I/O subsystems. To prevent UO subsystems from forever bottle-necking multiprocessors and limiting the range of feasible applications,new I/O subsystems must be designed.The successful design of computer systems (both hardware and software)depends on a thorough understanding of their intended use. A system'sdesigner optimizes the policies and mechanisms for the cases expected to bemost common in the user's workload. In the case of multiprocessor filesystems, however, designers have been forced to build file systems basedonly on speculation about how they would be used, extrapolating fromfile-system characterizations of general-purpose workloads on uniproces-sot and distributed systems or scientific workloads on vector supercom-puters (see sidebar on related work). To help these system designers, inJune 1993 we began the Charisma project, so named because the projectsought to characterize I/0 in scientific multiprocessor applications from a vari-ety of production parallel computing platforms and sites.The Charisma project is unique in recording individual read and writerequests4n live, multiprogrammmg, parallel workloads (rather than fromselected or nonparallel applications). In this article, we present the firstresults from the project: a characterization of the file-system workload onan iPSC/860 multiprocessor running production, parallel scientific appli-