Diurnal availability for peer- to-peer systems

Abstract

Ensuring content availability in a persistent manner is essential for providing any con- sistent service over peer-to-peer (P2P) systems. In this thesis we introduce an e cient protocol, called DATA (Diurnal Availability by Temporal Assemblage) of peers, to design highly available peer-to-peer systems irrespective of peer uptime and churn. Although peers in a P2P system join or leave at whim, it has been found that most of the peers follow diurnal pattern i.e. daily habit of availability governed by the time of day e ect. When considering a global P2P system, the cyclic behavior of peers situated at di erent time zones can be found complementary of one another. Our approach utilizes the diurnal pattern of globally dispersed peers to develop a grouping strategy. The objective of each group is to ensure 24x7 data availability within the group. In our approach, 24-hours of a day are divided into multiple time slots and DATA layer on each peer maintains its own historical probability of being online in each time slot. Using a gossip-based protocol, each peer discovers its best matching peers and groups with 4 to 6 peers in such a way that the combined availability in each slot within a group is close to 1. Simulation results reveal that our protocol converges fast and ensures high availability for each group with minimal overhead.