Multi-Epoch Scheduling Within the Real-Time Execution Performance Agent Framework Sam Siewert siewerts@thinker.colorado.edu Department of Computer Science University of Colorado, Boulder, CO 80309-0520 Gary Nutt nutt@cs.colorado.edu Department of Computer Science University of Colorado, Boulder, CO 80309-0430 Abstract In earlier work, we have described the Real-Time Execution Performance Agent (RT EPA) which accommodates mixed hard and soft real-time processing with measurable reliability by providing a confidence-based scheduling and execution fault handling framework. Based on experience with the RT EPA with a space telescope application, a theory was formed for determining scheduling feasibility for a set of services partitioned into multiple scheduling epochs, which are simultaneously active, yet mutually exclusive. This paper explains the multi-epoch theory whose utility has been demonstrated, but for which a formal implementation within the RT EPA framework remains to be completed. 1.0 Introduction The RT EPA is a framework for ongoing research for systems that support mixed hard and soft real-time services, especially systems which include processing pipelines between source and sink device interfaces . The RT EPA theoretical results include three major new real- time theories [Si00]: 1) An engineering view of real-time scheduling that inherently includes mixed hard and soft services with quantifiable reliability and confidence in the system. 2) Confidence-based thread admission and monitoring reducing RM (Rate Monotonic) pessimism in execution time bounds and providing reliable response prior to deadlines. 3) Evidence that thread sets can be admitted to multiple on-line epochs with priority changes between epochs, but fixed within an epoch. Research contributions 1 and 2 are discussed elsewhere [SiNu96], [SiNuHa99]. The third research contribution, multi-epoch scheduling, is described in this paper. Multi- epoch scheduling is orthogonal to the confidence-based scheduling features of the RT EPA, but was derived from experimentation with the RT EPA kernel monitoring technology as applied to the NASA SIRTF/MIPS (Space Infrared Telescope Facility / Multi-band Imaging Photometer for SIRTF) space telescope instrumentation processing [Si00b]. Without multi-epoch scheduling, the SIRTF/MIPS application would not provide reliable telescope exposure data processing and the science of the mission would have been adversely affected. This paper describes RT EPA multi epoch scheduling theory, presents preliminary results from the SIRTF/MIPS application, and discusses how the theory can formally be integrated into the existing RT EPA scheduling and execution monitoring framework. 2.0 Multi-Epoch Scheduling Theory and Related Research Traditional RM theory establishes the window of scheduling feasibility to be the longest overall period based upon the theorem formalized by Lehoczky, Sha, and Ding [BriRoy99]. In contrast, the goal of Pfair is to ensure that all tasks make progress at a steady rate, proportional to the weight of the task (utility) [Baruah97]. The Pfair algorithm reduces the window of scheduling feasibility to a window shorter than even the shortest release period. Multi-epoch scheduling lies between the Pfair extreme and RMA (though it is defined at a much higher level of granularity than either. MES, considers the possibility that multiple, independent chains of dependent real-time tasks – scheduling epochs – can be active at the same time, but only one of the epochs can be released at any given point in time (see Figure 1). This scenario is common to any system, which has mutually exclusive operational modes as the