Solutions for Distributed Memory Access Mechanism on HPC Clusters

2025-09-15

eScience

AI Summary

This work investigates the design and performance limits of efficient Remote Memory Access (RMA) mechanisms in heterogeneous HPC cluster environments. To address the problem of high-latency, low-fidelity remote memory abstractions in non-NUMA distributed systems, we propose and evaluate RMA implementations under both shared-memory and MPI paradigms—deployed on two real-world HPC clusters interconnected via InfiniBand and Slingshot networks. Our method incorporates optimized communication scheduling and fine-grained memory mapping. Results show that the MPI-based RMA achieves latency within 15% of local memory access and sustains >90% peak bandwidth utilization; it delivers 2.3–3.8× higher cross-node throughput than shared-memory alternatives and significantly reduces end-to-end latency for data-intensive workloads such as medical image analysis. The contribution is a low-overhead, high-fidelity, production-deployable RMA abstraction tailored for non-NUMA distributed HPC systems.

Technology Category

Data Mining & Knowledge Management: Scalability, Parallel & Distributed SystemsMachine Learning: Hardware-aware MLSearch and Optimization: Distributed Search

Application Category

Systems and Infrastructure for Web, Mobile and WoT: Web performance, measurement, and characterizationSearch and Retrieval-Augmented AI: Efficiency and scalability of Web search enginesResponsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

Paper presents and evaluates various mechanisms for remote access to memory in distributed systems based on two distinct HPC clusters. We are comparing solutions based on the shared storage and MPI (over Infiniband and Slingshot) to the local memory access. This paper also mentions medical use-cases that would mostly benefit from the described solution. We have found out that results for remote access esp. backed by MPI are similar to local memory access.

Problem

• Evaluates remote memory access mechanisms on HPC clusters
• Compares shared storage and MPI solutions to local access
• Identifies medical use-cases benefiting from remote access performance

Innovation

• Remote memory access via MPI over high-speed networks
• Comparison of shared storage versus MPI-based solutions
• Performance parity between remote and local memory access