Towards Predictable Execution of Safety-Critical Tasks on Mixed-Criticality Multi-Core Platforms | Consortium for Embedded Systems | SIU

Abstract: Multi-core architectures are a natural choice for integrating multiple independent functionalities into a single node in a cost- and space-effective manner. However, for systems with multiple levels of criticality, transporting highly safety-sensitive (HSS) applications (such as those for avionics systems with Design Assurance Level (DAL) of grade from DAL-C to DAL-A) onto a multi-core platform and sharing the benefits of the new computing environment with other less safety-sensitive (LSS) applications that have lower assurance levels but may be computation-intensive presents challenging problems in ensuring predictability/determinism of the HSS applications while still maintaining acceptable Quality of Service (QoS) for the LSS applications. The dominant approach towards isolating HSS and LSS tasks is the use of a virtualization environment (hypervisor) on top of the underlying multi-core platform. In prior work, extensive experimentation was conducted on the Freescale P4080 platform to characterize the behavior of HSS tasks in the presence of LSS tasks when executing them on different, statically derived partitions residing on separate cores. The goals of the proposed project are to apply the results of this characterization to end-use scenarios and to develop policies to exploit hardware mechanisms such as cache locking, cache partitioning and message passing among partitions to maintain determinism of HSS applications under regular and overload situations while maintaining QoS for the LSS applications.

Problem: Deterministic execution of HSS tasks in the presence of LSS tasks and other HSS tasks is challenging. The goal of this project is to develop policies to maintain responsiveness of HSS applications under regular and overload situations.

Rationale / Approach: The PIs propose to employ the following approach to achieve the goals of the proposed project: 1) Cache locking and partitioning are effective techniques to improve the predictability of HSS tasks and several policies for the same have been proposed. The PIs propose to explore the application of policies such as minimize utilization (Lock-MU) and minimize interferences (Lock-MI) [13] to end-use scenarios, identified by the workload characterization conducted in prior work. 2) Hypervisors typically allow configuration of one partition as a manager, giving this partition rights to pause and resume other partitions. The PIs propose to explore the use of such a manager partition a) to dynamically control the resource usage of LSS tasks under overload situations in an effort to maintain deterministic execution of HSS tasks and b) for power/energy conservation. Some of recently proposed techniques for energy saving scheduling in multicore systems as in [13-17] will be investigated and compared on the Freescale P4080 multi-core platform under a mix of HSS and LSS tasks from benchmark suites such as the MRTC WCET and EEMBC benchmarks.

Novelty: While cache locking, partitioning and partition management mechanisms are not new concepts, there is no study/research applying these in a safe manner to mixed-criticality workloads executing in virtualized environments on multi-core architectures.

Potential Member Company Benefits: The results of the proposed project will provide the basis for safe execution of mixed-criticality workloads on multi-core architectures with support for virtualization.

Deliverables for the proposed year: The deliverables for this project during the second year are as follows: 1) A comparative study of cache locking policies for given end-use scenarios on the Freescale P4080 platform. 2) A study of dynamic pausing/resuming of partitions (i.e., partition scheduling) to handle overload situations and for power/energy conservation. 3) Modified hypervisor and operating system source code, if any.

Milestones for the proposed year: Q1: Exploration of existing research in the area of cache locking and partitioning. Q2: Workload characterization and end-use scenario analysis under cache locking and partitioning schemes. Q3: a) exploration of mechanisms to create and configure manager partitions; b) development of strategies for dynamic resource management using manager partitions.