On scheduling atomic and composite multimedia objects

In multi user multimedia information systems e g video on demand news editing the pol icy employed to activate queued requests has a signi cant impact on the average startup latency observed by the users In this paper we view the retrieval of an object as a task and study the problem of scheduling such tasks In this scheduling problem tasks are IO bound and each may utilize multiple disks Each task acquires and releases disks in a regular manner based on the layout of its referenced object on the disks In addition there might be temporal relationships among multiple tasks that constitute a composite task While previous studies have focused on both scheduling subtasks localized to a single disk e g GSS YCK and determining the regular pattern of disk utilization per task e g striping BGMJ the method for selecting a task for activation has received little attention In this study we formalize a class of task scheduling problems that arise in a large class of multimedia applications We show that task scheduling problems are intractable NP hard justifying the use of heuristics We tailor some of the well known scheduling heuristics studied in other contexts and by means of a simulation study conclude that one of the heuristics FFD ECF is superior to the other alternatives We also demonstrate that the average latency time can be decreased further by employing bu ers to resolve contention This research was supported in part by the National Science Foundation under grants IRI IRI NYI award and CDA and a Hewlett Packard unrestricted cash equipment gift