Toward Broad-Spectrum Autonomic Management

Abstract

Abstract-The fields of autonomics and system configuration share a common goal in decreasing the cost of ownership of large fabrics. In this paper we present a combined vision in which the technical advances of autonomics and the usability advances of system configuration are merged. We present some early system configuration research that forms the first steps toward this vision.Computer fabrics 1 are becoming ever more complex and ever more central to the operation of organisations. The cost of a single fabric failure can easily dwarf the cost of employing its administrators [1] but, as for software developers before them [2], IT managers are discovering that simply adding more people to a complex problem does not necessarily provide better results.Autonomics has as its origin the attempt to improve the robustness of an infrastructure by enabling the fabric itself to respond to changes in its environment [3]. But how can we be sure that the the large-scale properties of the fabric will be maintained as it attempts to heal itself? Allowing the fabric to modify itself and evolve its own solutions will make management more difficult, not less, unless the power of administrators to understand and control their fabric is not also increased.The authors' background is not in autonomics, but in system configuration [4], the study of correctly, accurately and scalably building and maintaining computer fabrics. In this paper we present our vision of an autonomic fabric, a fabric which can recover from many errors without human intervention, yet never be beyond the control or understanding of its administrators. This, we believe, will be a fabric which is affordable, both in terms of the effort needed to maintain it and in terms of its support for critical applications.

Managing Real-World System Configurations with Constraints

“…Autonomics [18] offers the promise of selfconfiguring systems which will help to reduce these costs, but there is a considerable gap between the vision of a fully autonomic system and the practical tools currently available for configuring the infrastructure of a computing site [4]. Smith and Anderson argued in [23] that a modular approach with autonomic capability at several levels provides a route towards this vision. This paper describes a practical example of the modular approach; the high-level experimental configuration tool PoDIM [12] is interfaced to the the well-proven (but essentially low-level) configuration tool LCFG [7].…”