Dependability modelling of Jgroup/ARM

Abstract

In the later years, fault tolerant distributed systems have been applied to a variety of network internal and external services. Keeping distributed systems consistent and fault tolerant require management functionality. Jgroup/ARM is a java based prototype system which automates this management functionality for fault tolerant distributed systems. This thesis presents an evaluation of the dependability characteristics of the Jgroup/ARM system. Both static and dynamic modelling techniques are introduced, but the main focus is on the dynamic techniques; state-diagrams and Petri net models. Previous work is evaluated to nd an approach suitable for dependability modelling of Jgroup/ARM. A system delivery model for Jgroup/ARM is developed based on its functionality. The monitored subsystem is dened to include the hardware (processors), a given number of services in the system and the replication management functionality (ARM framework). A Petri net model of Jgroup/ARM is developed in the Möbius modelling tool based on the system delivery model. A model of a single service replicated on a cluster of processors is developed, analyzed and expanded to include multiple services and the ARM framework functionality. The dependability of Jgroup/ARM is evaluated through simulating example scenarios in Möbius. The results show that the system availability is very high, even with a relatively high failure rate.