A framework for statistical multiplexing onto a variable-bit rate output channel

This paper investigates soft multiplexing, an issue that arises in VP-based ATM networks. The main distinguishing feature of this form of multiplexing (as opposed to traditional statistical multiplexing) is that the output channel is a variable-bit rate VP. The concepts of work conservation and greediness were introduced to study soft multiplexing systematically. We assume that the overall multiplexed output traffic is regulated and policed by a leaky bucket. Before being allowed to access the output channel, each input traffic stream is also controlled by a leaky bucket. It is shown that although strict performance guarantee can be provided to each input traffic stream by independent leaky bucket operation, output-channel utilization can be rather low, and it can be improved significantly if we allow some interactions among the leaky buckets of the input streams. In particular, we show that two simple soft-multiplexing schemes, excess token passing and token borrowing, can achieve higher output-channel utilization, better delay and loss performance without sacrificing guarantee to each session