Collaborative environments range from the very primitive (IM) to a rich mixture of media, computation, and threedimensional graphics. Critical to all such systems is scalability : how many users can a system support in a single session? How many simultaneous sessions can be supported? These questions ultimately reduce to bandwidth and computation consumption of the collaborative system; the values, in turn, depend upon the architectural choices made in the implementation. Should the system be client-server or peer-to-peer? Should data or computation be replicated? Are there tradeoffs? Does it depend on the nature of the collaboration?

In this paper, we derive bounds on bandwidth and latency for consistent collaborative systems. A consistent system is one in which all participants agree on the order of events across all peers. We examine and calculate the bandwidth and latency bounds on various system architectures. We calculate a theoretical lower bound for computational and bandwidth complexity, and compare the architectural bounds to the lowest global bound.