A Software-Defined Approach for QoS Control in High-Performance Computing Storage Systems

TL;DR

This paper proposes a software-defined approach to ensure QoS in HPC storage systems, addressing the limitations of provisioning by enabling flexible, scalable, and efficient resource management without extensive reconfiguration.

Contribution

It introduces a novel software-defined method for QoS control in HPC storage, overcoming provisioning limitations and enhancing flexibility and scalability.

Findings

Improved QoS guarantees in HPC storage systems.

Reduced reconfiguration efforts through centralized control.

Enhanced flexibility and scalability of storage management.

Abstract

High-performance computing (HPC) storage systems become increasingly critical to scientific applications given the data-driven discovery paradigm shift. As a storage solution for large-scale HPC systems, dozens of applications share the same storage system, and will compete and can interfere with each other. Application interference can dramatically degrade the overall storage system performance. Therefore, developing a flexible and effective storage solution to assure a certain level of resources per application, i.e. the Quality-of-Service (QoS) support, is critical. One of the common solution to achieve QoS assurance for storage systems is using provisioning technique~\cite{3}. Provisioning refers to the ability of providing certain amount of resources for applications and expected workloads. However, provisioning has limitations such as requiring the detailed knowledge of the expected workloads. In addition, the storage workloads are transient hence expensive to be satisfied. Due to these limitations, providing QoS storage systems through provisioning is challenging. In this research, a software-defined approach~\cite{0} is proposed as a flexible solution to achieve QoS guarantee for storage systems. The driving force of using a software-defined approach instead of the traditional approaches, is that it has the ability to enable a more flexible, scalable, and efficient platform. For example, if any changes occurred in the system, it does not necessarily need to re-configure thousands of devices; instead, with re-configuring a logically centralized component, other devices will be automatically notified.