Special Issue on Management

This special issue contains seven papers originally presented at an International Workshop on Services for Managing Distributed Systems (SMDS'95), held in September 1995 in Karslruhe, Germany. The workshop was organized to present the results of two ESPRIT III funded projects, Sysman and IDSM, and more generally to bring together work in the area of distributed systems management. The workshop focused on the tools and techniques necessary for managing future large-scale, multi-organizational distributed systems. The open call for papers attracted a large number of submissions and the subsequent attendance at the workshop, which was larger than expected, clearly indicated that the topics addressed by the workshop were of considerable interest both to industry and academia. The papers selected for this special issue represent an excellent coverage of the issues addressed by the workshop.

A particular focus of the workshop was the need to help managers deal with the size and complexity of modern distributed systems by the provision of automated support. This automation must have two prime characteristics: it must provide a flexible management system which responds rapidly to changing organizational needs, and it must provide both human managers and automated management components with the information that they need, in a form which can be used for decision-making. These two characteristics define the two main themes of this special issue.

To satisfy the requirement for a flexible management system, workers in both industry and universities have turned to architectures which support policy directed management. In these architectures policy is explicitly represented and can be readily modified to meet changing requirements. The paper `Towards implementing policy-based systems management' by Meyer, Anstötz and Popien describes an approach whereby policy is enforced by event-triggered rules. Krause and Zimmermann in their paper `Implementing configuration management policies for distributed applications' present a system in which the configuration of the system in terms of its constituent components and their interconnections can be controlled by reconfiguration rules. Neumair and Wies in the paper `Case study: applying management policies to manage distributed queuing systems' examine how high-level policies can be transformed into practical and efficient implementations for the case of distributed job queuing systems. Koch and Krämer in `Rules and agents for automated management of distributed systems' describe the results of an experiment in using the software development environment Marvel to provide a rule based implementation of management policy. The paper by Jardin, `Supporting scalability and flexibility in a distributed management platform' reports on the experience of using a policy directed approach in the industrial strength TeMIP management platform.

Both human managers and automated management components rely on a comprehensive monitoring system to provide accurate and timely information on which decisions are made to modify the operation of a system. The monitoring service must deal with condensing and summarizing the vast amount of data available to produce the events of interest to the controlling components of the overall management system. The paper `Distributed intelligent monitoring and reporting facilities' by Pavlou, Mykoniatis and Sanchez describes a flexible monitoring system in which the monitoring agents themselves are policy directed. Their monitoring system has been implemented in the context of the OSIMIS management platform. Debski and Janas in `The SysMan monitoring service and its management environment' describe the overall SysMan management system architecture and then concentrate on how event processing and distribution is supported in that architecture.

The collection of papers gives a good overview of the current state of the art in distributed system management. It has reached a point at which a first generation of systems, based on policy representation within systems and automated monitoring systems, are coming into practical use. The papers also serve to identify many of the issues which are open research questions. In particular, as management systems increase in complexity, how far can we automate the refinement of high-level policies into implementations? How can we detect and resolve conflicts between policies? And how can monitoring services deal efficiently with ever-growing complexity and volume?

We wish to acknowledge the many contributors, besides the authors, who have made this issue possible: the anonymous reviewers who have done much to assure the quality of these papers, Morris Sloman and his Programme Committee who convened the Workshop, and Thomas Usländer and his team at the Fraunhofer Institute in Karlsruhe who acted as hosts.