Shared-buffer smoothing of variable bit-rate streams

Abstract

We study network servers that transmit variable bit-rate streams for real-time playback at remote clients. We introduce an algorithm that removes peaks of disk bandwidth by prefetching stored stream data into the shared buffer space of the server. Using a mathematical framework, we show that our algorithm has optimal smoothing effect to the server disk bandwidth over time. Emergence of inexpensive specialized devices makes prevalent the assumption of limited hardware resources for playback clients, and insufficient previous techniques that can only prefetch stream data into the client buffer space. We incorporate our algorithm into a prototype server, and demonstrate significant increase in the number of streams concurrently supported at different system scales. We also extend our algorithm to stripe variable bit-rate streams across heterogeneous disks. We achieve high bandwidth utilization across all the different disks, and improve the server throughput by several factors at high loads.