Peer Based Scalable Distributed Tuple Space

Abstract

With respect to temporal and spatial decoupling concept, coordination model like tuple space has proven to be well adopted and suitable for distributed peer-to-peer systems. MPI, publisher/subscriber, RPC and workflows are other proposed models for coordination activities in distributed systems. We have focused on decentralized structured tuple space model which identifies set of data separated from each other known as a tuple template. In our research we have used pastry structured overlay on which we built our framework. Since pastry is a DHT based implementation which is highly scalable as well as fault tolerant, tuple space middleware is developed on top of pastry and uses its predefined functions to join the network and routing for locating purpose. In this thesis we have discussed how distributed shared space model can address separation of concern and No SQL paradigm. Moreover, in our approach the middleware provides an interface to access tuple space in order to execute jobs using primitives defined as take(), read() and write() with features of scalability, latency for execution with increasing number of peers for same independent task and effect of tuple space model to massively parallelism tasks. Tuple space model provides developers to use its functionalities to perform and execute multiple jobs simultaneously using predefined function in pastry and framework developed. We describe our approach, along with the design of a prototype implementation and its quantitative evaluation.