Jini Technology in Grid System Development

Juhász Zoltán, Dr. <juhasz@irt.vein.hu>

Veszprémi Egyetem

Andics Árpád <?>

Veszprémi Egyetem

Póta Szabolcs <?>

Veszprémi Egyetem

The past decade has brought dramatic changes in the use of parallel computers. Due to technological and economical reasons, parallel computers made up from commodity (off-the-shelf) components have replaced the propriety architectures. Fast networking technology has made the use of local, then later national, clusters feasible for parallel computing. Today, it is possible to connect geographically widely disperse computers to form large virtual parallel systems, known as grids. However, the development, programming and operation of these systems is not without problems.

The operation of grid systems is fundamentally different from that of traditional parallel computers. The large distances and the presence of computer networks result in errors – processors can fail, computers can be switched off, the network can go down, etc. Moreover, the computers in the system can include different types of processors and operating systems. Therefore, grid systems must cope with operation errors, software and hardware heterogeneity, and should provide user management facilities for remote machines.

Jini Technology developed By Sun Microsystems in 1999 makes it possible to connect digital devices into an ad hoc network automatically, without human intervention. This federation of devices and services is dynamic, participants can come and go as they please and use the network when they need it. Jini builds on the Java language environment and creates a fault-tolerant, secure and platform independent distributed object system that makes the connection and operation of networked services as simple as possible.

The presentation will report on the results of our current research project. We describe our experience with Jini and overview the architecture, operation and programming technology of the JM (Jini Metacomputing) system we have developed. This system makes it possible to connect diverse computers into a federation on demand basis to form a large computing resource pool for clients. Finally, we describe our results of running the system and test programs on a local university cluster as well as on a system created from computers located at different Hungarian universities.