COMBINING HARDWARE MANAGEMENT WITH MIXED-CRITICALITY PROVISIONING IN MULTICORE REAL-TIME SYSTEMS

Safety-critical applications in cyber-physical domains such as avionics and automotive systems require strict timing constraints because loss of life or severe financial repercussions may occur if they fail to produce correct outputs at the right moment. We call such systems “real-time systems.” When designing a real-time system, a multicore platform would be desirable to use because such platforms have advantages in size, weight, and power constraints, especially in embedded systems. However, the multicore revolution is having limited impact in safety-critical application domains. A key reason is the “one-out-of-m” problem: when validating real-time constraints on an m-core platform, excessive analysis pessimism can effectively negate the processing capacity of the additional m-1 cores so that only “one core’s worth” of capacity is available. The root of this problem is that shared hardware resources are not predictably managed. Two approaches have been investigated previously to address this problem: mixed-criticality analysis, which provision less-critical software components less pessimistically, and hardware-management techniques, which make the underlying platform itself more predictable. The goal of the research presented in this dissertation is to combine both approaches to reduce the capacity loss caused by contention for shared hardware resources in multicore platforms. Towards that goal, fundamentally new criticality-cognizant hardware-management tradeoffs must be explored. Such tradeoffs are investigated in the context of a new variant of a mixed-criticality framework, called MC2, that supports configurable criticality-based hardware management. This framework allows specific DRAM banks and areas of the last-level cache to be allocated to certain groups of tasks to provide criticality-aware isolation. MC2 is further extended to support the sharing of memory locations, which is required to realize the ability to support real-world workloads. We evaluate the impact of combining mixed-criticality provisioning and hardware-management techniques with both micro-benchmark experiments and schedulability studies. In our micro-benchmark experiments, we evaluate each hardware-management technique and consider tradeoffs that arise when applying them together. The effectiveness of the overall framework in resolving such tradeoffs is investigated via largescale overhead-aware schedulability studies. Our results demonstrate that mixed-criticality analysis and hardware-management techniques can be much more effective when applied together instead of alone.Doctor of Philosoph