Video-on-Demand (VoD) is a service that allows people to request and view stored videos or movies of their choice directly online from a VoD service provider. The selected streaming videos are then delivered over the broadband Internet. The bursty nature of Variable-Bit-Rate (VBR) compressed video (such as MPEG) poses some important issues for video delivery over high-speed networks due to its significant bit rate variation over multiple time scales. However, sufficient quality of service (QoS) mechanisms must be in place before it can be widely enabled and deployed over Internet.
Conventionally a static bandwidth level close to the peak rate is reserved for a streaming video flow. Any static allocation of network resources for VBR video traffic would be difficult and inefficient considering the peak rate to be significantly higher than the average data rate. Since the traffic pattern over time is already known for pre-recorded videos, this issue is addressed by the Renegotiated Constant Bit Rate (RCBR) service which proposes QoS allocation over multiple time scales. Since this mechanism has been tested via simulations and analysis only we implemented it on a real test bed with a VoD server and clients to study its performance. We observed that under heavy bandwidth constraints the performance of RCBR is much better than traditional CBR in terms of packet loss rate. We also implement a new Adaptive Buffer Window mechanism and the concept of application level smoothing to increase the scalability of a VoD server.
