Coeus: An Approach for Building Self-Learning and Self-Managing Systems

Systems are becoming exceedingly complex to manage. There is an increasing trend towards developing systems that are self-managing. The eventual target is to build systems in which the administrator specifies overall goals and policies while the system internally decides how these can be achieved. To achieve this target, we need to automate the administrator’s task of tuning configuration parameters and invoking of system services such as back-up, replication and so on. The task of automation is non-trivial because of two reasons. First, administrators use implicit heuristics to decide how the system is tuned. Second, by monitoring the impact of their previous decisions, administrators can refine their heuristics i.e. self-learning. Policy-based infrastructures have been used to provide a limited degree of automation, by mapping actions to specific system-states. But, there is no systematic approach or mechanism within this infrastructure to support automation of the administrator’s tasks mentioned above namely decision-making and self-learning. This paper proposes an approach to solve this problem by using Coeus: a model and framework for building self-managing systems. The key contributions of this work are:

• Describes a model to help administrators and system-builders specify policies that can capture the implicit management heuristics.
• Using the Coeus model, it describes automation of the tasks of decision-making and self-learning.
• Describes the Coeus framework in the context of existing policy-based management infrastructures.