An adaptive scheduling scheme for serving VBR-encoded multimedia streams

We propose a novel scheduling scheme, in which a new user is admitted for service by a multimedia server only if the service requirements of all the existing users and the new user can be satisfied. Each user is assumed to be requesting variable bit rate (VBR) MPEG2 encoded media stream. Based on the service requirements, the users are classified into two classes: (1) deterministic service users, and (2) statistical service users. Under the deterministic service, a user gets his data continuously without any loss of media information. However, for statistical service users, occasional loss of media data (i.e., B-frames) is possible though the loss ratio is tightly bounded. The B-frames of the statistical service users are dropped only during the overflow cycles, i.e., when a data retrieval cycle is estimated to be skipping the deadline of that round. Since the actual cycle time may be different from the estimated cycle time, a cycle may be estimated to be an overflow cycle but it may actually turn out to be an underflow cycle. Similarly, the vice-versa is also possible. To deal with this uncertainty, we introduce a novel concept which is based on a quantity δ. The quantity δ is the measurement of the maximum error possible between an estimated and an actual cycle length. In our proposed scheme, a cycle is considered an overflow cycle only if the estimated cycle is overflowing by more than δ, otherwise, the cycle is an underflow cycle, and no frame is dropped. In this scenario, a deadline can be missed by a maximum of 2δ duration. To overcome this problem, we suggest that the playback is delayed by 2δ duration right from the beginning so that the users will not experience any playback discontinuity. Our scheme is much simpler than the schemes suggested in the literature and results in better utilization of the resources. Extensive simulations have been conducted to demonstrate the effectiveness of this scheduling scheme