Designing a Fault–tolerant Jini Compute Server

Java-based tuplespaces provide a simple infrastructure for scientific distributed computing. There are several classes of problems that are not efficiently solvable in JavaSpaces model of computation while efficiently solvable in other tuplespace implementation. JavaSpaces can be used for high performance computing if viewed less strictly in the heritage of Linda and more as a platform-neutral code delivery mechanism. This paper presents an early design of JavaSpaces compute server. We describe how we use mobile co-ordination and agent wills to provide faulttolerance in Jini based compute servers. Preliminary experimental results show performance gains made when mobile co-ordination is used. In this paper we apply the mobile co-ordination and agent wills [12] to provide fault-tolerance in compute servers based on tuple spaces. The first section the tuple space paradigm and the Linda model for parallel computation. The next section provides an overview of Jini and JavaSpaces [14]. The third section describes in detail our design and then we present our conclusions. 1. The Tuple Space Paradigm Linda is a well known co-ordination model [3]. The fundamental concepts of Linda are tuples, templates and tuple spaces. A tuple is an ordered collection of fields. Each field has a type and a value associated to it. A field with both a value and a type is known as an actual. The tuple < 1989int, “Linda”string , 1.0real > is a tuple containing three fields, with the type of the field shown as a subscript of the value. Tuples are placed into tuple spaces and are removed from tuple spaces using an associative matching process. A template is similar to a tuple except the fields do not need to have values associated to them, but all fields must have a type. A field that has only a type and no value is known as a formal. A template is a tuple which can have formals. A template matches a tuple if they have the same number of fields, and 2000 Mathematics Subject Classification. 65Y05, 68Q85. 1998 CR Categories and Descriptors. F.1.2 [Computation by Abstract Devices]: Modes of Computation – Parallelism and concurrency; G.4 [Mathematics of Computing]: Mathematical Software – D.1.3 [Programming Techniques]: Concurrent Programming – E.1 [Data]: Data Structures – Distributed data structures.