I2S Cluster Overview

An HPC (High Performance Computing) cluster is a collection of interconnected computers, sometimes called nodes, that can work independently or together as a unified system. Nodes are connected through high-speed, low-latency network fabrics like InfiniBand or RoCE, allowing them to perform parallel processing where complex tasks are divided into smaller parts and distributed across the nodes simultaneously. The nodes also often share filesystems. In the case of the I2S cluster, /home, /work, and /apps are all NFS filesystems shared by every node. /scratch is located on a high-performance distributed parallel filesystem called Lustre which we typically mount over InfiniBand to maximize bandwidth and minimize latency.

Ways to utilize HPC clusters:

1. You can use a cluster like a large, unified machine if you make use of MPI implementations like OpenMPI and MPICH.
2. You can use it to quickly churn through a large queue of small tasks asynchronously
3. You can use individual nodes as if they were remote workstations.

Cluster node types

This is an auto-updated summary of nodes in the I2S cluster.