Sustainability
High Performance Computing is energy-intensive, and running the cluster efficiently keeps both its environmental footprint and its running costs down. This page collects Anunna's approach to sustainable operation.
Buying hardware
When buying new hardware we take power use into account. It is part of the scoring matrix, as is packing of the hardware.
Powering down idle nodes to save energy
Compute nodes that are not in use are powered off automatically to save energy, and powered back on when the scheduler needs them. You may see this reflected as a powered-off state when inspecting cluster usage (see node usage).
Water usage/liquid cooling
As the HPC is rather power hungry, this means that the electricity that is used in the end will be converted into heat. That heat needs to be removed from the machines, and transported to the outside air.
For most of the machines the CPUs are liquid cooled, this is a lot more energy efficient than cooling it by passing cooler air over them. Water has a higher specific heat capacity, and can thus hold more heat per volume than air. So one needs to pump a lot less water around, than one would do using air.
There is still air cooling needed for the rest of the components, but this saves a lot of energy.
As the room where the servers live has to be closed of from the outside, the hot air and hot water is passed through a secondary liquid cooling system, which is itself cooled with outside air.
All the liquid cooling loops (internal and external) are closed, we do not use external water to cool.
Using resources efficiently
The most effective thing users can do is to request only the resources a job actually needs and no more — over-requesting CPUs, memory, or wall time ties up hardware that could otherwise be powered down or used by others. See Choosing a node (constraints) for sizing jobs appropriately.