Probabilistic dropping in push and pull dissemination over distributed hash tables

Dynamic information management via Distributed Hash Tables (DHT) is an important problem which revolves around the trade-off between data freshness and the overhead due to information updates. We propose two different algorithms based on information pull and information push models, that enable dynamic information dissemination with low overhead over a DHT. We exploit the concept of popularity of specific items, which is evaluated by performing a real-time analysis of the query distribution, and allows to decrease a significant fraction of messages without impairing the query resolution process. We have measured the overhead savings and compared the performance of the two approaches by extensive simulations using real workload traces.