MAGIKMAIDS: Mobile agents for in-kernel monitoring, assessment and intervention in distributed systems

Abstract

As distributed systems have begun to play a more and more central role in modern computing, their size and complexity has increased up to the point where centralized, human-mediated management of the system is unfeasible. A need arises to endow these systems with a form of distributed intelligence so as to allow scalable self-organization and timely response to changing conditions. Mobile agents are one well-established method of implementing such versatile distributed intelligence and many mobile agent platforms have been created for use in computer networks and distributed systems. While all such efforts so far have been limited to user space, we consider the novel approach of implementing such a platform in kernel space. We begin by making a case for why a kernel space implementation would present worthwhile advantages, then we investigate possible implementation paradigms and follow with the architecture and details specific to our chosen approach. Finally, we evaluate our platform in terms of performance and ease of use and draw the appropriate conclusions.