The main goal of CUP is to continuously harvest the benefits of PCX. In doing so, there are two key performance questions to address. First, by how much does CUP reduce the average query latency? Second, how much overhead does CUP incur in providing this reduction?
We first define the notion of a CUP tree. We use this definition to present a cost model based on economic incentive used by each node to determine when to cut off the propagation of updates for a particular key. We give a simple analysis of how the cost per query is reduced (or eliminated) through CUP. We then describe our experimental results comparing the performance of CUP with that of PCX.