Skip to Main Content
Data stored in the P2P data grids are replicated at peers to achieve good query latency, low communication cost, and high availability. However, replication also raises consistency issues, which should be well defined and carefully maintained to facilitate better user collaboration. Epidemic algorithms were designed to satisfy a level of consistency weaker than serializability but preserve the causal order of update operations or transactions. However, with large system scale and with high access concurrency, epidemic lazy update algorithms can also introduce system delusion (e.g., with too many conflicting update or read transactions, system cannot reach consistency that there is no way to repair by reconciling conflicting transactions), which is due to their lazy update nature. Also, in P2P systems, some peers may be malicious because of being compromised, thus, security is a critical issue that should be addressed. In this paper, we propose an update mechanism to limit system delusion while providing high access concurrency. A two-level vector clock is developed to maintain the causal order of transactions in the data grid and maintain the consistency of peers with transaction logs. Security protocol is designed and integrated into the update mechanism to address the security issues.
Date of Conference: 14-16 Nov. 2007