Towards robust self-managed systems

Abstract

Software is pervasive. It plays a part in every aspect of our lives, forming a significant part of every automated system or tool that we use, from washing machines to healthcare systems. Like the washing machine, some of these applications are simple, selfcontained devices; however the majority are far more complex. They are generally distributed systems which rely on interacting, distributed subsystems of software components to perform their function. They are expected to interact with users with changing needs and with other systems with unreliable service provision. The challenge is to provide these software systems in such a way that they are robust in the presence of major issues such as change and complexity. Change is inherent, both in the changing needs of users and in the changes which take place in the operational environment of the system. Hence it is essential that our systems can adapt as necessary to continue to achieve their goals. Change is also induced by failures or the unavailability of parts of the system. It is therefore necessary to envisage dynamically changing configurations of software components so as to adapt to the current situation. Dynamic change, which occurs while the system is operational, requires that the system evolves dynamically, and that the adaptation occurs at run-time. Complexity requires that we use rigorous techniques to design, build and analyse our software and thereby avoid unnecessary design flaws. This implies the need for analytical techniques which cope with changing goals and the changing compositions of adaptive software. Since the complexity and response times required by the changes may not permit human intervention, we must plan for automated management of change. The systems themselves must be capable of determining