Lustre File Systems

The Lustre file systems are planned to become the main shared file system for HPC.


Each group has a limited amount of space which is enforced by the file system. If this limit is exceeded, the whole group will not be able to write new data.

You can check the current use with:

% lfs quota /lustre/hpc
Please, read 'man lfs' for more info - or the online documentation

A note about quota: Once you (or your group) get close to/above your allocated quota, then allocating more space will be really slow on Lustre.


The groups that have purchased backup, can access the backups on the frontends using this path,

% /shared/backup/<group>
where group is your group name. Each subdirectory name here is a date of the format YYYY-MM-DD, i.e. 2017-05-16. This is the date that the backup was started.

The amount of backup space is the same as the group quota for the Lustre file system. Therefore, there will be at least one complete backup. Now, the backup is compressed harder than the Lustre file system, so there will likely be space for some history (days of backup). Only the difference between one backup till the next will be saved. If there is enough space we will keep 365 days of backup. When space becomes tight we will delete the oldest, but keep 3 monthly and 3 weekly backups. If that is not enough, the monthly and then weekly backups will be sacrificed.

Partial backup

A few groups have partial backup of the lustre file system. They need to opt in to get files backed up.

This is done by changing the group that the file belongs to. So i.e. if you are in the foo group, you can include a file in the backup by calling:

% chown :foobac <file>
This way we know which files should be backed. You can change all files/dirs in a directory by:
% chown -R :foobac <dir>

For convenience you can make a directory where all files created will belong to the backup group.

% chown :foobac <dir>
% chmod g+s <dir>
Note. If you move a file or directory (mv) to this directory it will not change group. So be careful!

The main reason we need this special backup group is to make it possible for Lustre to track the limited partial backup space. Quota is only accounted for on a per group or user basis. Soon Lustre will also have project quotas (quota per directory sub-tree), which will be a better option.

To check how much backup space it left run:

% lfs quota -h -g foobac /lustre/hpc/

Notes about sshfs (eg. ERDA)

In order to use sshfs, the permissions for group needs to have execute access to the the path. You need to verify/correct this, if you get:

$ fusermount -u sshfs/
fusermount: failed to chdir to /lustre/hpc/group/user: Permission denied

The simple way is to do:

cd ..
chmod g+x $USER

If you get:

ls: cannot access sshfs: Transport endpoint is not connected
try running:
fusermount -u sshfs