High performance computing (HPC) refers to aggregated computing power. This grouping of computers can be referred to as a cluster, and the individual computers in a cluster are nodes.
A node has most of the same parts as a desktop or laptop computer. It has memory/RAM, storage, and a processor/CPU (or processors). However, a node is usually more powerful than a standard desktop or laptop. The following table compares my desktop, a Mac mini purchased in early 2021, to an Atlas compute node.
| Specification | Mac mini | Atlas Compute Node |
|---|---|---|
| Processor Cores | 8 | 48 |
| Graphics Processing Cores | 8 | |
| Memory (gigabytes) | 16 | 384 |
| Storage (gigabytes) | 500 | 2000 (local, plus access to a shared filesystem) |
As our tools generate more and more data, the time and resources needed to analyze the data grow larger. Sometimes, completing an analysis on a desktop or laptop is possible but takes too long. In other cases, the size of the data might be larger than the resources available on the machine.
In the simplest case, running some analysis on a single Atlas node offers several benefits over running the same analysis on the average desktop.
When the data can be split into discrete chunks (individual samples in an experiment), each chunk can be analyzed on a separate node to parallelize the analysis. When the data can’t be split into chunks, certain programs can use the resources of multiple nodes as though running on a single large node.