PERFORMANCE OF INTER-MEDIA SYNCHRONIZATION IN DISTRIBUTED AND HETEROGENEOUS MULTIMEDIA SYSTEMS

Citation
Pv. Rangan et al., PERFORMANCE OF INTER-MEDIA SYNCHRONIZATION IN DISTRIBUTED AND HETEROGENEOUS MULTIMEDIA SYSTEMS, Computer networks and ISDN systems, 27(4), 1995, pp. 549-565
Citations number
19
Categorie Soggetti
Computer Sciences","System Science",Telecommunications,"Engineering, Eletrical & Electronic","Computer Science Information Systems
ISSN journal
01697552
Volume
27
Issue
4
Year of publication
1995
Pages
549 - 565
Database
ISI
SICI code
0169-7552(1995)27:4<549:POISID>2.0.ZU;2-5
Abstract
Future multimedia environments are expected to be distributed and hete rogeneous. In such environments, media sites ranging from sophisticate d workstations to simple media capture and display subsystems such as ISDN videophones and audiophones (together referred to as Mediaphones) that lack the capability to run clock synchronization algorithms will be connected directly to the network. Towards such environments, we d evelop mechanisms and protocols for synchronous access to multimedia s ervices over integrated networks. In the inter-media synchronization t echnique we present, for facilitating synchronous retrieval, a multime dia server computes at the time of recording of a multimedia object a range of relative temporal positions or stamps (called an RTS interval ) for each media unit that it receives from mediaphones. During playba ck, the multimedia server detects RTS mismatches between media with th e help of feedback messages transmitted back by the mediaphones, and s teers the mediaphones back to synchrony. We propose predictive policie s for resynchronization and compare their performance for video/audio playback with other resynchronization policies, such as conservative, aggressive, and probabilistic. These performance studies reveal that p redictive policies perform uniformly well at all levels of asynchrony, unlike their conservative and aggressive counterparts. Moreover, in m ost cases, predictive policies match probabilistic policies in effecti veness, without imposing the same computational demands as the probabi listic policies.