Analysis and Implementation of Global Preemptive Fixed-Priority Scheduling with Dynamic Cache Allocation

Abstract

We introduce gFPca, a cache-aware global pre-emptive fixed-priority (FP) scheduling algorithm with dynamic cache allocation for multicore systems, and we present its analysis and implementation. We introduce a new overhead-aware analysis that integrates several novel ideas to safely and tightly account for the cache overhead. Our evaluation shows that the proposed overhead-accounting approach is highly accurate, and that gFPca improves the schedulability of cache-intensive tasksets substantially compared to the cache-agnostic global FP algorithm. Our evaluation also shows that gFPca outperforms the existing cache-aware non-preemptive global FP algorithm in most cases. Through our implementation and empirical evaluation, we demonstrate the feasibility of cache-aware global scheduling with dynamic cache allocation and highlight scenarios in which gFPca is especially useful in practice. Abstract-We introduce gFPca, a cache-aware global preemptive fixed-priority (FP) scheduling algorithm with dynamic cache allocation for multicore systems, and we present its analysis and implementation. We show that a naïve extension of existing overhead analysis techniques can lead to unsafe results, and we introduce a new overhead-aware analysis that integrates several novel ideas to safely and tightly account for the cache overhead. Our evaluation shows that the proposed overheadaccounting approach is highly accurate, and that gFPca not only improves schedulability of cache-intensive tasksets substantially compared to the cache-agnostic global FP algorithm but also outperforms the existing cache-aware non-preemptive global FP algorithm in most cases. Through our implementation and empirical evaluation, we demonstrate the feasibility of cacheaware global scheduling with dynamic cache allocation and highlight scenarios in which gFPca is especially useful in practice.

Holistic Resource Allocation for Multicore Real-Time Systems

“…Cache resource. Several cache partitioning techniques have been proposed to reduce the shared cache interference [9,21,22,26,28,59,61,62,68,74]. The software-based approach reorganizes a task's memory layout to allocate a specific cache area to the task using, e.g., page coloring [28,39,67] or compiler-based [42] techniques.…”

“…The software-based approach reorganizes a task's memory layout to allocate a specific cache area to the task using, e.g., page coloring [28,39,67] or compiler-based [42] techniques. The hardware-based approach [39,61] leverages the cache partitioning capability in recent COTS processors, such as the Cache Allocation Technology (CAT) in Intel processors [5] and the Lockdown-by-Master (LbM) technology in ARM processors [2]. We use the hardware-based approach (specifically, Intel's CAT) to achieve the per-core cache allocation as it is much more efficient than the software-based approach.…”

vCAT: Dynamic Cache Management Using CAT Virtualization

“…Software-based approaches have been used to provide both static [14,21] and dynamic [25,29] allocations. Previous work has also explored hardware-based approaches to dynamically allocate cache partitions to tasks, e.g., using the PL310 cache controller [27] or the Intel's CAT [32]. Some techniques also combine software and hardware approaches to support finer-granularity partitions [8,16].…”

“…It reserves a small (configurable) number of partitions to be shared among all best-effort tasks, and uses all the rest for real-time tasks. We implemented two strategies for allocation: the first allocates partitions to tasks in a first-come-first-served basis, as they are scheduled on the VCPUs; and the second gives priority to a more critical task, i.e., allows it to preempt lower-criticality tasks to acquire sufficient partitions, similar to the approach used in [27]. In both cases, if the task's execOnFewer flag is set, and if the VM has some but fewer than the requested number, the kernel simply allocates the available partitions to the task (and let it execute) to maximize core utilization and to minimize preemption overhead.…”

Cache-aware interfaces for compositional real-time systems

“…In our prior work, we have explored this approach for real-time components [9], and our evaluation results show that it can help improve schedulability substantially compared to cache-agnostic scheduling. In the following, we discuss two important research problems and potential directions towards this approach for compositional systems.…”