Space and compute requirements of PPM for migration within the secondary storage

PPM algorithm is used in this work to dynamically migrate and reorganize data in a three-tiered secondary storage hierarchy implemented using software RAID. The three levels implemented in the prototype named temperature sensitive storage (TSS) are RAID 10, RAID 5 and compressed RAID 5 levels. Our earlier study confirmed the usefulness of this algorithm in reducing performance degradation with compressed RAID 5 levels being 20% or more of the working-set size. The usefulness of PPM for such an application is gauged not only by using performance metrics such as throughput, residual compressed RAID 5 migrations, hit and miss ratios at different migration boundaries, but also by its space and compute requirements. Results using two file-system traces Sitar and Harp show that the worst case storage requirements of the PPM has been well below 10 MB and CPU usage has been moderate. The total migrations generated for 400000 accesses has been less than 3% of the total accesses seen by the TSS device driver.