META-LEVEL CONTROL IN DISTRIBUTED REAL-TIME SYSTEMS
Krithi Ramamritham, John A. Stankovic, Wei Zhao · International Conference on Distributed Computing Systems · 1987
In distributed systems where the system state and task characteristics change dynamically, no single scheduling algorithm performs well in all situations. Ideally, the system should be able to use the algorithm(s) needed for a particular situation depending on the state of the nodes and the communication network as well as the task characteristics. What we are proposing is to control the scheduling. We term this higher-level control as meta-level control. In this paper we discuss how meta-level control can improve the performance and enhance the adaptability of resource allocation schemes. We illustrate our ideas by studying meta-level control for distributed hard real-time systems. In this context, meta-level control can be used for the following: Selecting the algorithm(s) used for scheduling tasks on a node, selecting the algorithms for cooperation among nodes, and selecting the values of scheduling parameters used in the chosen algorithm(s). After a discussion of considerations underlying such selection, we focus on the problem of implementing meta-level control in such a way that the overhead inherent in such control does not negate the performance improvement possible via meta-level control. Our preliminary studies show that if meta-level control is exercised judiciously, significant performance improvement in system performance will result.