#A Robust Protocol for Aggregation in Dynamic Peer-to-Peer Networks
Aggregation in peer-to-peer networks provides a global view of system properties, such as average computational load and total free disk space, which can be used for administering the system. Among the various aggregation protocols, gossip-based aggregation protocols provide faster aggregation of the node values in a scalable and adaptive manner. This is particularly important in peer-to-peer networks which are essentially dynamic, due to node failures, node churn, communication delays and communication failures. These dynamics affect the correctness of the system aggregate, which is a function of the individual node values (node contributions).
We first propose a protocol hierarchy for gossip-based aggregation that provides the capability to perform accurate estimation of the system aggregate, even in the presence of node failures. In this protocol hierarchy, we develop a robustness protocol for aggregation in dynamic peer-to-peer networks that ensures the correctness of the system aggregate amidst node failures. Our protocol detects node failures and performs the necessary cleanup operations. The evaluation of our protocol shows that the error of the converged aggregate from the system aggregate is very low. We observe that the failure detection time is independent of the network size and the number of nodes that fail. Additionally, we investigate the factors affecting the accuracy of a node's estimate of the system aggregate (node estimate) and introduce a metric called the 'confidence value' which gives an indication of this accuracy at runtime. Hence, our robustness protocol enables accurate computation of system aggregate even in the presence of node failures.