What kind of self-aware systems does the Grid need

In this paper we draw on computer science, biology, psychology, and philosophy to examine what sort of self-awareness the Grid needs. The vision of the Grid is of an enormously large-scale heterogeneous distributed computing system that enables the transparent sharing of geographically distributed resources, including hardware, software, and data. Several influential Grid researchers have said that the Grid requires self-aware systems, without being explicit about what “self-aware” means. We will identify properties – automatic resource allocation, self-configuration, robustness, fault handling and remediation, and security that we believe are necessary for the Grid. We outline several definitions of self-awareness from philosophy, cognitive science and computer science. These fall into two families; the first family is concerned with the external behaviour of the system, the second with its internal processes and structure. The commonest computer science definition of self-aware systems is in the second family, and involves the property of reflection [Baclawski, 2000] the ability of a program to manipulate as data something representing the state of the program during its own execution. Some philosophers and cognitive scientists suggest a similar definition [Anderson, 1996]. The main theme of this paper is that reflection is in fact not necessary for the Grid. We will show, with examples, that it is possible to have systems that exhibit properties that we have identified as being necessary for the Grid, without using reflection. We will argue that using reflection – or indeed, any self-awareness property concerned with the internal processes of the system in the design of Grid systems might be inefficient. We will conclude that Grid systems should use selfawareness properties based on the external behaviour of the system, or no selfawareness at all, depending on the task, and will discuss the types of tasks for which self-awareness is most appropriate. Some of this investigation may be useful to philosophers. By looking practically at what properties are needed for a system to operate effectively within a complex, dynamic environment, without assuming that this system is biological, we may indicate an interesting avenue of exploration, investigating whether living creatures – and in particular humans – themselves have the properties that we identify. Moreover, if we show that some properties of self-aware living creatures are not necessary for the Grid, this raises the question why these properties arise in living creatures. Finally, our analysis that self-awareness is useful for some tasks but should be avoided for others may provide confirmation for some observations in cognitive science.