A compartmentalised supercomputer

Even though a supercomputer usually runs some variant of Linux, it does work differently from a regular Linux workstation or server. When you log on on a PC, you immediately get access to the full hardware. This is not the case on a supercomputer. A supercomputer consists of many compartments:

• When users log on to a supercomputer, they land on the so-called login nodes (called user access nodes on some HPE computers). These are one or more servers that each look like a regular Linux machine but should not be used for big computations. They are used to prepare jobs for the supercomputer: small programs that tell the supercomputer what to do and how to do it.

• Each supercomputer also has a management section. This consists of a number of servers that are not accessible to regular users. The management section is responsible for controlling and managing the operation of the supercomputer, including deciding when a job may start and properly starting that job.

• Each supercomputer also has a shared storage section, a part of the hardware that provides permanent storage for data or a scratch space that can be used on the complete machine.

• But the most important part of each supercomputer is of course the compute section, or compute sections in many cases as most supercomputers provide different types of compute resources to cover different needs of applications.

In these notes, we will mostly talk about the structure of the compute section of the cluster, but we will also cover some typical properties of the storage system as that has a large influence on what one can do and what one cannot do on a supercomputer.