Storage – General

Local Scratch

Every system has a local partition available for writing to, this is /scratch.

For typical grid submission (via ARC) when your jobs run you should already drop into a sub-directory within /scratch/<jobid>, if you happen to do direct submission (without ARC) then please ensure you use /scratch/$USERNAME.

We recommend using scratch as much as possible as this will always be faster than any shared network storage.

Scratch space is not shared and is only available locally, so if you require the data over many systems, please ensure you copy this to and from a shared system prior/end of your work.

Scratch is never backed up and is liable to be erased as soon as a system is faulty.

Shared Network

Home – Grid Users

Your home space is on a shared network server and is automatically mapped to you ( ~/, /mt/home/$USERNAME and /mt/data-grid/$USERNAME ).

By default all users get 20GB, we can expand this upon request to 50GB or 100GB. It is backed up daily.

The home server is used by everyone all the time, so we request that you do not run batch jobs, jupyter notebooks or any significant IO heavy work from/to your home directory.

Home – Fielding Users

Your home space is on a shared network server and is automatically mapped to you ( ~/ and /mt/fielding-home/$USERNAME).

By default all users get 100GB, there is no room for expansion. It is backed up daily.

The home server is used by everyone all the time, so we request that you limit your job usage of /mt/fielding-home otherwise it can cause significant performance issue for all users and nodes, please push your IO to /scratch and then copy/rsync back to home at the start/end of each job (and any convenient checkpoint locations).