[Paper Review] OpenMosix approach to build scalable HPC farms with an easy management infrastructure

R. Esposito, P. Mastroserio+2 more|ArXiv.org|May 26, 2003

Distributed and Parallel Computing Systems5 citations

TL;DR

This paper presents OpenMosix as a scalable, low-maintenance solution for building high-performance computing (HPC) clusters in high-energy physics (HEP) environments. By leveraging diskless nodes and network booting, it enables efficient load balancing and automatic process migration, significantly boosting system performance with minimal administrative overhead.

ABSTRACT

The OpenMosix approach is a good solution to build powerful and scalable computing farms. Furthermore an easy management infrastructure is implemented using diskless nodes and network boot procedures. In HENP environment, the choice of OpenMosix has been proven to be an optimal solution to give a general performance boost on implemented systems thanks to its load balancing and process migration features. In this poster we give an overview of the activities, carried out by our computing center, concerning the installation, management, monitoring, and usage of HPC Linux clusters running OpenMosix.

Motivation & Objective

• Address the challenge of managing large-scale HPC clusters in high-energy physics (HEP) with minimal administrative overhead.
• Develop a scalable HPC infrastructure that maintains high performance under dynamic workloads.
• Implement an easy management solution using diskless nodes and network boot procedures to reduce system complexity.
• Evaluate the effectiveness of OpenMosix in real-world HEP computing environments for performance and manageability.
• Demonstrate the feasibility of using OpenMosix for load balancing and process migration in production HPC clusters.

Proposed method

• Deploy Linux clusters using diskless nodes that boot over the network, eliminating local storage and simplifying maintenance.
• Utilize OpenMosix's built-in load balancing to distribute computational workloads across available nodes dynamically.
• Enable transparent process migration between nodes to balance system load and improve resource utilization.
• Implement centralized monitoring and management tools to oversee cluster operations and performance.
• Integrate the cluster into existing HEP computing workflows to support real-time data processing and analysis.
• Leverage standard networking protocols and PXE (Preboot Execution Environment) for reliable and automated node booting.

Experimental results

Research questions

• RQ1 Can OpenMosix provide effective load balancing and process migration in a production HPC environment for high-energy physics?
• RQ2 How does the use of diskless nodes and network booting impact system scalability and administrative overhead?
• RQ3 To what extent does OpenMosix improve overall system performance in HEP computing workloads?
• RQ4 What are the practical challenges and benefits of deploying OpenMosix in a real HEP computing center?
• RQ5 How does OpenMosix compare to traditional HPC cluster management in terms of ease of deployment and maintenance?

Key findings

• OpenMosix enabled the creation of a scalable HPC cluster with minimal administrative effort due to the use of diskless nodes and network booting.
• The system achieved significant performance improvements through automatic load balancing and dynamic process migration across nodes.
• The management infrastructure was simplified, reducing hardware and maintenance costs compared to traditional disk-based clusters.
• The cluster demonstrated robust performance in real HEP computing workloads, validating OpenMosix as a viable solution for HEP environments.
• The deployment was successfully integrated into the computing center’s workflow, supporting ongoing data processing tasks.
• The solution proved effective for both batch and interactive workloads, showing adaptability across diverse HEP computing needs.