Real-Time Scheduling with Hardware Data Structures

Two essential features of a real-time operating system (RTOS) are time management and task scheduling. Such features reduce software developers’ burden of designing, implementing, and validating generic system infrastructure, thus lowering costs and decreasing application time-to-market. However, there is a cost that is often paid as system overhead during the runtime. Hardware coprocessors that encapsulate RTOS services can reduce system overheads and increase the amount of CPU time available to applications. Prior work in scheduling coprocessors have moved scheduling and event processing of a RTOS into hardware. Our work returns the control logic of scheduling coprocessors back to software, and captures the data-centric logic as a hardware data structure. Separating the control and data aspects of scheduling coprocessors yields efficient yet flexible hardware support for real-time systems. We demonstrate the flexibility of the hardware data structure by implementing two classic periodic task schedulers, the rate monotonic (RM) and earliest deadline first (EDF) algorithms, and use the same structure for managing timers.