Compositional Schedulability Analysis of Multicore Modular Avionic Architectures
Journal of Computers
This paper presents a compositional schedulability analysis of multicore modular avionic systems (IMA). It provides a fine grained description of the software architecture and behavior to reduce the over-approximation, and a holistic analysis to check schedulability while considering computation requirements and shared memory interference. The system is structured in terms of subsystems, encapsulating ARINC653 partitions, each of which runs on a processing core. The schedulability analysis is
... rformed for each subsystem individually while accounting for the memory interference that would result if the core under analysis runs effectively alongside with the rest of cores. Thereafter, we introduce an architecting technique to relocate functions between tasks located at the same partition in case of non schedulability, to derive potential schedulable system configurations delivering the same functionality. Schedulability is formally analyzed using Uppaal model checker. Our evaluation results show that our compositional analysis technique consumes up to 95% less than regular analysis, in terms of analysis time and memory space. 1202 perfect memory access i.e., the memory is immediately available whenever an access request occurs (hits) -  . Such an estimation of the WCET at task level is most likely under-approximating. We believe that estimating the execution time at a lower level of the granularity (e.g. function level) results in less underapproximation of the execution time. Thus, the more granular the behavior representation is, the more optimistic the WCET will be. The under-approximation difference between task and function levels could be non comparable and would increase drastically with fine grained description levels. The interference resulting from shared memories and communication means (buses, networks) is a determinant factor in the schedulability of component-based real-time system. The interference time is in fact related to the number of concurrent components and the bandwidth of shared resources. Accordingly, accurate schedulability techniques require a holistic analysis where both computation and memory interference/ communication have to be considered together ,  . A surge of progress has been achieved in the area of schedulability analysis of multicore systems through an intensive use of model-based settings and formal methods. However, due to the systems size in avionics, given by the integration of huge number of concurrent applications, the use of formal methods can end up in state space explosion. Different techniques have been considered to bypass the state space explosion and provide upper bound guarantees on the schedulability, we cite abstraction-based , compositional analysis  and incremental analysis  . In this paper, we introduce a model-based framework for fine grained modeling and formal schedulability analysis of multicore avionic (IMA-driven) systems with shared memories. The system is structured in terms of subsystems, encapsulating ARINC653 partitions, each of which runs on a processing core. The schedulability analysis is performed for each subsystem individually while accounting for the memory interference that would result if the core under analysis runs effectively alongside with the rest of cores. Our framework does not require the WCET of application tasks to be provided in order to check the schedulability, but rather it assumes that the individual instructions (low level functions) of the application have been identified and associated both an execution time and a memory access pattern. In case of non-schedulability of a subsystem, we provide a technique to re-engineer the faulty subsystem by relocating functions between tasks located within the same partition, calculate what will be the resulting load and analyze the underlying schedulability. We restrict the migration of functions between tasks within the same core only to maintain the partition-based design concept and functional architectural constraints of IMA. The rest of the paper is organized as follows: Section II cites the most relevant related work. Section III describes the IMA architecture and the overall scheduling and analysis methodology introduced in this paper. Section IV describes our formal modeling of modular avionic systems. Section V describes our compositional schedulability analysis and re-location techniques. Section VI discusses the scalability of our compositional analysis compared to regular schedulability analysis. Section VII concludes the paper. Related Work Analyzing the schedulability of a system application while considering an abstraction of the execution platform leads necessarily to underestimate the system requirements (workload) in terms of resources. This causes certainly serious deficits during deployment where processes run much longer than what they should, due to interference and non-deterministic contention of shared memories and communication means. Many researchers have been paying considerable attention to the schedulability and memory interference of multicore systems , , , -. Boudjadar et al  introduced a compositional framework for the schedulability analysis of uniprocessor hierarchical scheduling systems. The system components analyzed individually are given with interfaces describing the resource budget (in terms of processor utilization time) that the component under Ph.D degree in December 2012 from Toulouse University France. His research interests include software architectures, model-based design and formal verification of embedded real-time systems. Boudjadar's research aims to develop advanced architecture description and analysis techniques for embedded real-time systems.