Adaptive control of prefetching

Redundant Arrays of Inexpensive/Independent Disks (RAID) remain the mass storage medium of choice for real-time processing and information storage and retrieval. This dissertation derives new adaptive pre-fetching algorithms for RAID systems using “life-time function analysis of programs.” The life-time function approach derives an inequality which allows us to determine when pre-fetching can be effective. This inequality relates the changes in disk transfer times to the changes in inter-page fault times with the ratio of disk transfer time to page fault time. We then use this inequality to design an adaptive pre-fetching strategy and to propose a practical implementation. Simulation experiments to evaluate the effectiveness of the proposed approach are carried out. The analytical approach based on a single process execution model is extended to a multiprogramming environment using a queuing network model, as well as to a simpler approach using a “parallel processing” approximation. Both synthetic and real trace driven simulations are conducted to confirm the significant increase in performance that can be achieved with our approach.