This chapter analyzes the performance of a multi-disk CM server utilizing the deadline-driven scheduling and random data placement (DD). Section 4.1 investigates the distribution of block retrieval times and its impact on the hiccup probability. We propose a taxonomy of deadline-driven approaches in Section 4.2. Section 4.2.1 introduces a technique to reduce both the hiccup probability and the startup latency. We evaluate and compare our proposed techniques in Section 4.3.
3.147.81.154