Scaling Analysis

  Any multicast protocol that makes decisions at a source on behalf of all the receivers in the network is fundamentally prone to scaling problems. There is simply no scalable mechanism to keep a tightly consistent view of every receiver in the network at each source. We explicitly address this scaling problem in SCUBA by sacrificing absolute consistency in favor of scalability through the use of sampling. In this section, we examine this tradeoff in detail, outline the parameters that we can adjust to tune SCUBA's scalability, and identify the fundamental limitations of our scheme.

There are two principal protocol parameters that limit the scalability of SCUBA with respect to session size:

• the control message size, and
• the expected source weight convergence time.

The size of each control message depends on the number of reported sources. If each receiver includes a weight for every active source, then the report size scales linearly with the number of active sources. In a large session with large numbers of active sources, the resulting message size will quickly exceed a reasonable limit, but we can avoid this problem by simply imposing a fixed limit on the message size. If we limit the number of sources for which each receiver reports to a small constant, then the control packet size is bounded. We claim that this limit is reasonable since human attention can realistically encompass only a handful of sources simultaneously.

The scalability analysis of the average source weight calculation time is more involved and we present it in the following section.

• Convergence Time