Enabling Scientific Workflow Scheduling Research in Non-Uniform Memory Access Architectures

2025-11-24

AI Summary

Data-intensive scientific workflows suffer from inefficient scheduling on NUMA-based HPC systems due to high variance in memory access latency and complex task–data placement constraints. Method: We propose nFlows—the first workflow runtime system supporting NUMA-aware modeling and direct execution on real hardware. It integrates NUMA-aware task scheduling, fine-grained memory binding, heterogeneous memory management (HBM/DRAM), and co-scheduling of accelerators (GPU/FPGA/NIC) with device affinity. It enables seamless validation of scheduling algorithms across simulation and physical platforms while deeply characterizing memory locality and data movement patterns. Results: Experiments demonstrate that nFlows significantly reduces cross-NUMA-domain data access latency, accurately identifies performance bottlenecks, and validates the critical impact of NUMA-aware co-optimization on memory-intensive workflows—achieving substantial end-to-end performance gains.

Technology Category

• Machine Learning: Hardware-aware ML
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Data Mining & Knowledge Management: Scalability, Parallel & Distributed Systems

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Data management and stream processing for Web, mobile and wireless applications
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Data-intensive scientific workflows increasingly rely on high-performance computing (HPC) systems, complementing traditional Grid and Cloud platforms. However, workflow scheduling on HPC infrastructures remains challenging due to the prevalence of non-uniform memory access (NUMA) architectures. These systems require schedulers to account for data locality not only across distributed environments but also within each node. Modern HPC nodes integrate multiple NUMA domains and heterogeneous memory regions, such as high-bandwidth memory (HBM) and DRAM, and frequently attach accelerators (GPUs or FPGAs) and network interface cards (NICs) to specific NUMA nodes. This design increases the variability of data-access latency and complicates the placement of both tasks and data. Despite these constraints, most workflow scheduling strategies were originally developed for Grid or Cloud environments and rarely incorporate NUMA-aware considerations. To address this gap, this work introduces nFlows, a NUMA-aware Workflow Execution Runtime System that enables the modeling, bare-metal execution, simulation, and validation of scheduling algorithms for data-intensive workflows on NUMA-based HPC systems. The system's design, implementation, and validation methodology are presented. nFlows supports the construction of simulation models and their direct execution on physical systems, enabling studies of NUMA effects on scheduling, the design of NUMA-aware algorithms, the analysis of data-movement behavior, the identification of performance bottlenecks, and the exploration of in-memory workflow execution.

Problem

• Scheduling scientific workflows on NUMA HPC systems with data locality challenges
• Addressing variable data-access latency in heterogeneous memory architectures
• Developing NUMA-aware scheduling algorithms for data-intensive workflow execution

Innovation

• NUMA-aware workflow execution runtime system
• Modeling and simulation for NUMA-based scheduling
• Bare-metal execution with data locality optimization