NERSC consists of multiple machines and services, each of which are useful for different tasks.
Although you can ssh directly to NERSC machines, you will likely have an unpleasant typing experience and may be interrupted by dropped connections, etc. A much better technique is to download the “NX client” software. This provides a remote desktop (KDE, just like the late 1990’s!) and allows you to disconnect / reconnect without losing work and time. Follow instructions here.
Within the NX client settings, I suggest clicking on the “Display” options and choosing:
These will ensure that you have a small amount of data being sent back and forth over the network, but still have minimal pixelization artifacts. Such settings usually work well even when tethering to a mobile phone for example.
This sytem has 2 types of nodes:
This system has 5500 nodes of older, traditional CPUs, with 2x12 core chips per node. Good for large jobs that use code which does not run well on the cori KNL nodes.
All users have a small home directory for storing things like scripts, git checkouts, etc. This space is not purged, but is fairly slow. Do not read or write large data files here.
Each individual user has 20TB of personal “scratch” storage. This location is different on each machine but can always be found in the “$SCRATCH” environment variable. This disk is very fast, but is purged after some weeks. Copy any data you want to save to a project directory or your home directory. You can see your current disk usage with:
%> myquota
Each project has shared storage in /project/projectdirs/(project name), which is typically 20TB, but can be increased if we fill that up. This filesystem is not purged. You can see current project disk usage with:
%> prjquota <project name>
More details on filesystems are here.
If you are transferring large amounts of data to or from NERSC, and / or backing that data up to HPSS (multi-petabyte tape system), then you should carefully read more about these services: