OptMRT: Near-Optimal Partitioning for Mixed Real-Time Tasks on Heterogeneous Architectures

View at publisher Summarise with VIRAT
Authors: Yuankai Xu, Yinchen Ni, An Zou, Yehan Ma
Cited by: 0
References: 30

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

To address the demands of modern artificial intelligence (AI) applications, heterogeneous architectures have gained widespread adoption, ranging from embedded devices to high-performance computing systems. These applications often involve tasks with timing constraints, which are commonly categorized into hard real-time (HRT) and soft real-time (SRT) domains. HRT tasks require strict deadline guarantees, whereas SRT tasks can tolerate occasional deadline misses. The coexistence of HRT and SRT tasks is prevalent in multi-functional systems like autonomous driving. However, scheduling such mixed real-time workloads on heterogeneous architectures remains challenging due to the divergent timing objectives between the two task types, the non-preemptive execution model on GPUs that aggravates blocking, and the CPU-GPU alternating execution inherent to heterogeneous platforms. To address these challenges, OptMRT (Near-Optimal Partitioning for Mixed Real-Time Tasks) introduces a bi-partition mechanism, incorporating a dual-mode mapping policy, an Streaming Multiprocessors (SM) allocation strategy guided by a unified timing metric, and a scheduling strategy tailored for mixed HRT and SRT workloads. Leveraging end-to-end response time analysis, OptMRT achieves near-optimal performance for both HRT schedulability and SRT timing performance. Extensive experiments show substantial improvements in schedulability, up to 2.25× for HRT and 2.33× for SRT, over state-of-the-art approaches.

Fields of Study & Subjects

Real-Time Systems Scheduling Embedded Systems Design Techniques Parallel Computing and Optimization Techniques

Publication details

Journal / source
IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
Volume
45
Issue
11
Pages
6147–6160
Publication date
25 March 2026
Language
EN