Computing systems have evolved from single machines in large machine rooms to millions of interconnected devices whose interactions create complex webs built on increasingly complex architectures consisting of multitudes of powerful devices running tens of millions of lines of code. The increase in size and complexity of interconnected heterogeneous systems has led to a similar increase in the cost and complexity of configuring and operating such systems. Complexity is not just a function of the number of devices, but also of diversity in their types and capabilities. Lastly, increasing globalization tends to lead to geographic dispersion of some large systems, which adds another dimension to this already complex scenario.

IBM highlighted this growing complexity and coined the term autonomic computing [1] to describe an approach for managing complexity that relies on designing and building computing systems capable of "managing themselves." Basically rooted in control theory, autonomic computing is a design principle and an architectural philosophy that advocates increased autonomy of computing nodes and self-management at the edge of the network.

It is, however, a challenge to build closed-loop managed systems in an open architecture where a system is composed of multiple components (hardware and software) from different vendors. This is because components need to interoperate seamlessly and have a shared common understanding and representation of policies, managed objects, and system events. At each level of the hierarchical management loop, components will run a closed autonomic loop with well-defined interfaces to communicate with their peers and with other levels of management hierarchy. In this paper, we explore the various standards that facilitate such an interaction, including WS-Management and the emerging model-based approaches.

This paper is organized as follows. Firstly, we briefly outline the conceptual architecture for autonomic computing. We then describe the multiple standards for enabling autonomic computing. Finally, we provide additional details on external interfaces for autonomic computing elements as they enable interoperability between autonomic managers in heterogeneous systems.