Parallel Program/Component Adaptivity Management

Grid computing platforms require to handle dynamic behaviour of computing resources within complex parallel applications. We introduce a formalization of adaptive behaviour that separates the abstract model of the application from the implementation design. We exemplify the abstract adaptation schema on two applications, and we show how two quite different approaches to adaptivity, the ASSIST environment and the AFPAC framework, easily map to this common schema. 1 An Abstract Schema for Adaptation With the advent of more and more complex and dynamic distributed architectures, such as Computational Grids, growing attention has to be paid to the effects of dynamicity on running programs. Even assuming a perfect initial mapping of an application over the computing resources, choices made can be impaired by many factors: load of the used machines and network available bandwidth may vary, nodes can disappear due to network problems, user requirements may change. To properly handle all these situations, as well as the implicitly dynamic behaviour of several algorithms, adaptivity management code has to be built into the parallel/distributed application. In so doing, a tradeoff must be settled between the complexity of adding dynamicity-handling code to the application and the gain in efficiency we obtain. The need to handle adaptivity has been already addressed in several projects (AppLeS [5], GrADS [10], PCL [8], ProActive [4]). These works focus on several aspects of reconfiguration, e.g. adaptation techniques (GrADS, PCL, ProActive), strategies to decide reconfigurations (GrADS), and how to modify the application configuration to optimize the running application (AppLes, GrADS, PCL). In these projects concrete problems posed by adaptivity have been faced, but little investigation has been done on common abstractions and methodology [9]. In this work we discuss, at a very high level of abstraction, a general model of the activities we need to perform to handle adaptivity in parallel and distributed programs. Our model is abstract with respect to the implemented adaptation techniques, monitoring infrastructure and reconfiguration strategy; in this way we can uncover the common aspects that have to be addressed when developing a This research work is carried out under the FP6 Network of Excellence CoreGRID funded by the European Commission (Contract IST-2002004265).