MULTIMEDIA STORAGE SERVERS - A TUTORIAL

Citation
Dj. Gemmell et al., MULTIMEDIA STORAGE SERVERS - A TUTORIAL, Computer, 28(5), 1995, pp. 40-49
Citations number
12
Categorie Soggetti
Computer Sciences","Computer Science Hardware & Architecture","Computer Science Software Graphycs Programming
Journal title
ISSN journal
00189162
Volume
28
Issue
5
Year of publication
1995
Pages
40 - 49
Database
ISI
SICI code
0018-9162(1995)28:5<40:MSS-AT>2.0.ZU;2-#
Abstract
On-line access to multimedia information--like books, periodicals, ima ges, video clips, and scientific data--is both possible and cost-effec tive, thanks to recent advances in computing and communication. Some m edia, such as audio and video, are classified as continuous because au dio samples and video frames, for example, have meaning only when pres ented in time. The design of multimedia servers thus fundamentally dif fers from conventional servers as a result of (1) real-time storage an d retrieval requirements, as well as (2) large storage space and data transfer-rate requirements of digital multimedia. In this tutorial, th e authors highlight the issues involved in meeting these requirements. For example, the critical components in the design of multimedia serv ices are storage servers that support continuous media storage and ret rieval, and network subsystems that synchronously deliver media inform ation, on time, to the client sites. In their survey of design issues, the authors present disk-scheduling algorithms (optimized for retriev ing multimedia information) for real-time recording and playback. The authors also discuss admission control algorithms that let a multimedi a server determine whether new services can be added without the serve r's violating the real-time requirements of existing ones. In terms of service, the authors assume that performance requirements of multimed ia storage servers include meeting all real-time deadlines, although s ome applications can tolerate missing one occasionally. Servers have s everal quality-of-service (QoS) categories: (1) Deterministic--All dea dlines are guaranteed to be met; (2) Statistical--Deadlines are guaran teed to be met with a certain probability (perhaps 90 percent); and (3 ) Background--No guarantees are given for meeting deadlines (the serve r schedules access only when there is time left over after all other s ervice). The authors also examine techniques for efficiently placing m edia information on individual disks, large disk arrays, and storage d evice hierarchies. Finally, they describe data structures that a multi media file system must maintain to facilitate random access and effici ent editing.