ABSTRACT

Carnegie Mellon University Technical Report CMU-CS-01-105, March, 2001.

PASTENSE: A Fast Start-up Algorithm for Scalable Video Libraries

Stavros Harizopoulos and Garth A. Gibson

School of Computer Science
Carnegie Mellon University
Pittsburgh, PA 15213

http://www.pdl.cmu.edu/

Striping video clip data over many physical resources (typically disk drives) balances video server load with less data replication. Current striped video delivery algorithms can have high start-up latency if the load is high. We propose a new, fast start-up algorithm, PASTENSE. This algorithm minimizes start-up latency by using aggressive prefetching to exploit disk idle time, and using available RAM to dynamically optimize the newly requested video's schedule. Our proposed method (a) does not require changes in the existing striped data placement (b) it never performs worse than alternate designs and (c) it achieves significant benefits: up to 9 times faster start-up times for high loads.

FULL PAPER: pdf / postscript