The Flatiron Institute, New York, a community of scientists using modern computational tools to advance the basic sciences, is deploying a 320-node addition to its research cluster that will be managed by cluster management software from Bright Computing, maker of Linux cluster automation and management platform for HPC and machine learning.
The implementation at Flatiron, the in-house research division of the Simons Foundation, will increase the institute’s clustered infrastructure to more than 1,300 nodes, all of it managed by Bright Cluster Manager. The platform balances and manages heavy workloads – including data analysis, theory, modeling and simulation – across such scientific disciplines as astrophysics, quantum physics, data science, mathematics and biology.
“When we first started out, the Flatiron Institute had very little staff,” said Flatiron’s Scientific Computing Core Co-Director Ian Fisk, Ph.D. “As we began to transition into a larger organization, we had to figure out how to manage a growing set of resources — but still with that limited staff. Bright Cluster Manager offered us a fully supported cluster management solution that allowed our team to manage our compute resources more efficiently with very little overhead. We first deployed Bright at the Flatiron Institute when we grew from a 16-node cluster to 120 nodes, and Bright has been managing the system ever since.”
The Bright Cluster Manager is designed to maximize resource utilization while supporting a heterogeneous compute environment, freeing researchers to focus on computational research rather than the complexities of cluster management, according to Bright. The software allows system administrators to provision hardware, operating systems and workload managers as a single unified infrastructure.
The implementation at Flatiron will increase the institute’s clustered infrastructure to more than 1,300 nodes, all of it managed by Bright Cluster Manager.
At Flatiron, Bright integrates hardware from a variety of hardware vendors, enabling staff to re-image the cluster for upgrades, change packaging, update configurations, allocate resources, and add nodes on the fly. Bright also has comprehensive monitoring and GPU management capabilities.
“The Flatiron Institute is comprised of over 200 scientists performing research in a variety of scientific domains,” said Martijn de Vries, CTO of Bright Computing. “This places a requirement on their clustered infrastructure to be computationally proficient in many different ways as the resources needed for these scientific domains can be quite different from discipline to discipline.
“Bright Cluster Manager’s hardware-agnostic platform provides the Flatiron Institute with a single unified system that reduces complexity and promotes efficiency and ease-of-use,” de Vries said. “We are dedicated to providing cluster management solutions to organizations such as the Flatiron Institute, that take the guesswork out of system administration and give researchers a faster and more efficient path to successful research.”
