The SRP Resource Sharing Protocol for Self-Suspending Tasks

Abstract

Motivated by the increasingly wide adoption of realtimeworkload with self-suspending behaviors, and the relevanceof mechanisms to handle mutually-exclusive shared resources, thispaper takes a new look at locking protocols for self-suspendingtasks under uniprocessor fixed-priority scheduling. Pitfalls whenintegrating the widelyadopted Stack Resource Policy (SRP) withself-suspending tasks are firstly illustrated, and then a new fine grainedSRP analysis is presented. Next, a new locking protocol,named SRP-SS, is proposed to overcome the limitations of theoriginal SRP. The SRP-SS is a generalization of the SRP to copewith the specificities of selfsuspending tasks. It therefore reducesto the SRP under some configurations and hence theoreticallydominates the SRP. It also ensures backward compatibility forapplications developed specifically for the SRP. The SRP-SScomes with its own schedulability analysis and configurationalgorithm. The performances of the SRP and SRP-SS are finallystudied by means of schedulability experiments.

On the QNX IPC: Assessing Predictability for Local and Distributed Real-Time Systems

“…The dependency can be broken by assuming that all jobs are killed after the deadline: this allows initially setting the WCRT bounds R j to the deadline D j . In this way, if threads are schedulable, no jobs are killed, making the assumption not necessary (more details in [16]). Iterative improvements are possible by running the analysis multiple times for all tasks, each time obtaining new (smaller) bounds for R i leveraging R j (see Eq.…”

“…Iterative improvements are possible by running the analysis multiple times for all tasks, each time obtaining new (smaller) bounds for R i leveraging R j (see Eq. ( 1)), updating the values of each R i with the newly computed R i , and reiterating the process until no improvement is achieved between two consecutive runs of the analysis [16]. Finally, note that the equation of Section 4.2.3 in [9] has been extended in Eq.…”

“…( 5) requires a preexisting bound on the response time of high-priority clients c j ∈ hep cl k (s h ), which can initially be upper-bounded with its deadline D j , and possibly refined in an iterative fashion as discussed in Section IV-A. This technique, well-known in literature (e.g., see [9,16]- [19]), is used multiple times in the following: the corresponding pre-existing WCRT bounds are denoted with the symbol R to distinguish from those are under derivation (denoted by R).…”

Bounding Memory Access Times in Multi-Accelerator Architectures on FPGA SoCs

“…When bounding response times of real-time tasks, it is common to assume that the interfering tasks complete by their deadlines to get rid of circular dependencies that arise in response-time equations. Please refer to [19] (Sec. VI.C) for further details about the validity of this approach.…”