BloomBox: Improving Availability and Efficiency in Geographic Hash Tables

Abstract

Mobile Ad Hoc Networks are important today for scenarios in which centralized cloud infrastructure is missing, has broken down, or imposes censure or undesirable monitoring of storage and communication. Unfortunately, existing peer-to-peer storage systems such as a Geographic Hash Table (GHT) can consume a significant amount of network bandwidth just to maintain a certain required number of replicas of the data due to the churn present in the network. The replicas have to continuously exchange heartbeat messages in order to detect failures of replicas. If heartbeat messages get lost, an unnecessary but expensive recovery protocol ends up wasting significant bandwidth. To avoid this, replicas are placed close to one another, but that makes them vulnerable to dependent failures.Based on Mergeable Bloom Filters, a new data structure proposed for peer-to-peer distributed systems, we build BloomBox, a failure detection protocol for a GHT. Our simulations show that BloomBox can significantly reduce bandwidth usage needed for regenerated blocks compared to heartbeat-based failure detection. Moreover, BloomBox can provide significantly better availability than protocols based on heartbeats by placing replicas in geographically diverse locations.