Data sets in large applications such as databases and data-mining applications are often too large to fit into a computer's internal memory. The resulting I/O (input/output) communications between internal memory and the slower external medium (such as disks) can present a major performance bottleneck. Among the possible computer architectural solutions to this problem are cluster computing platforms, which are collections of non-dedicated systems (for example, PCs) based on low cost...
Presents recent theoretical and practical advances that will provide new and experienced users a greater understanding of the subject of I/O management. Also includes five research studies contributed by specialist researchers in the field.
Foreword13Introduction15Ch. 1Motivating I/O Problems and their Solutions23Ch. 2Parallel Sorting on Heterogeneous Clusters63Ch. 3A Sensitivity Study of Parallel I/O under PVFS93Ch. 4Effective I/O Bandwidth Benchmark107Ch. 5Parallel Join Algorithms on Clusters133Ch. 6Server-side Scheduling157Ch. 7A Large Virtual NVRAM Cache for Software RAID183App. 1Matrix Product MPI-2 Codes211App. 2Selected Web Sites Related to I/O235Index239