A robust high-performance time partitioning algorithm: the digital engine operating system (DEOS) approach

RTOSs supporting critical avionics applications must provide time partitioning; that is, the ability to guarantee a certain amount of execution time to applications in one time partition regardless of resource requests from applications in a different time partition. Time partitioning permits applications of differing criticalities to be cohosted on a single processor (e.g., DO-178B levels A-E). This paper presents the major design components of slack stealing with its supporting time partitioning algorithms. These algorithms have been implemented in DEOS, the RTOS developed for use in Primus Epic, Honeywell's next generation integrated avionics system. From a time partitioning perspective, DEOS differs from earlier ARINC 653 systems in that it uses preemptive fixed priority (PFP) scheduling, and consequently the time partitioning model includes preemption. Additionally, DEOS supports dynamic thread and dynamic time partition allocation. With the slack and time partitioning upgrades added to DEOS, several improvements were observed. About a threefold increase in communication throughput between a processor and a remote file server and approximately a sevenfold reduction in the amount of a priori reserved execution time required for "fast enough" response times of noncritical applications were reported. Also, certain display tasks were able to achieve higher average refresh update rates using slack. Our algorithms provide robust time partitioning capabilities, enabling the safe cohosting of COTS software with safety critical software without decreases in the COTS software performance.