Distributed management of network links supporting multimedia computing

The paper analyses distributed algorithms used to control network links carrying random data traffic generated by multimedia computing based on local information. Distributed computing, multimedia mail, and interactive shopping are just a few examples of applications where the current approaches for traffic control (as foreseen by ITU-TS) are not suitable. Traffic descriptors (other than maybe peak rates) are difficult to determine for the user. Global knowledge of the network state is most of the time obsolete due to latency, and therefore approaches must be imagined which do not rely on such information. The paper analyses the options available for bandwidth management in ATM-based networks carrying predominantly data traffic, develops a rationale for supporting “fill-in” approaches capable of exploiting very high bandwidth links and quantifies the trade-off associated with these approaches, using both trace driven and event driven simulation