Several broadcasting schemes are proposed to distribute media on isochronous channels. The server bandwidth requirement is the main concern in broadcasting. Other trade-offs include the required bandwidth and buffer space at the client side and the startup latency. Periodic broadcasting schemes are server-centered; they give high priority to saving server bandwidth. Bandwidth skimming schemes are client-centered; they care about reducing client bandwidth and startup latency.
The main idea of periodic broadcasting is to partition each video file into K segments, which are broadcast repeatedly on separate logical channels. The intuition is that the first segment should be short and repeated with a high frequency so the startup latency is low; while the first segment is being received and played, the time can be used to wait for or receive the second segment so the second segment can be longer and repeated with a lower frequency; since the second segment is longer, it takes longer time to receive and play it; more time is gained for the third segment so it can be even longer and repeated less frequently; and so on. Long segments are repeated with low frequency so the bandwidth is saved.
The sizes of the K segments are a increasing series, but they may increase in a different way. How to design the increasing series is the core of periodic broadcasting schemes. It effects how the server bandwidth is allocated, the startup latency, and the client bandwidth and buffer space. The four schemes mentioned in this group each have its own way to decide this increasing series. The Pyramid Broadcasting and the Permutaion-Based Pyramid Broadcasting use a geometric series with a adjustable parameter. The Skyscraper Broadcasting uses a recursive function to generate the series, which grows slower than the geometric series. The Reliable Periodic Broadcasting adopts an algorithm with the maximum possible segment size increase, which results in a Fibonacce series in some case. A special feature of Reliable Periodic Broadcasting is that each segment is delivered by an approximation of a digital fountain to achieve loss recovery without retransmission.
Bandwidth skimming is a client merging technique, in which a new client listens to its own new stream as well as the most recently initiated on-going stream that it can merge with. This means the client needs a small skim of extra bandwidth to catch up a stream close to it. This skim of client bandwidth can substantially reduce server and network bandwidth. The skim can be tuned to provide trade-offs between playback quality and required server bandwidth.
The solution for streaming multimedia over the Internet proposed by the Hofmann paper is new caching techniques, including segmentation of streaming objects, dynamic caching, and self-organizing cooperative caching. The authors integrate these techniques and implement a complete caching architecture called SOCCER. Their simulations show that SOCCER can significantly reduce network and server load compared to existing caching schemes.