ERT: Data placement based on estimated response time for P2P storage systems

Abstract

Storage system using P2P architecture is an alternative to traditional client-server system that offers better scalability and fault tolerance while eliminating the single point of failure. P2P storage systems with heterogeneous nodes face performance and scalability challenges, one of which is contributed by their data placement method. Existing data placement methods, such as the implementation of the Kademlia protocol, store data at the closest node, where the distance is measured by bitwise XOR operation between identifiers of data and the node. This approach is highly scalable because it does not require global knowledge for placing and retrieving the data. It does not however consider the heterogeneous performance of the nodes, which can result in imbalanced resource usage. Other works implement criteria-based node selection that addresses the heterogeneity of nodes. However, such approaches often cause subsequent data retrieval to require global knowledge of the data location. This paper introduces a novel data placement method called Estimated Response Time-based (ERT), which stores data to selected nodes based on dynamic estimation of nodes response time. ERT models the data placement process using queueing theory. Besides, it incorporates the predicted response time, alongside other criteria, for node selection decision-making. The experimental results in a real P2P storage system indicate that ERT reduces the standard deviation of response time of nodes by 17.57% compared to the native Kademlia, 39.01% compared Kademlia with Virtual Node, and by 49.24% compared to Throughput-based data placement. Further, relying on a modified DHT data structure of Kademlia allows ERT to eliminate the need for global knowledge during data retrieval.