Marvin has a large Lustre storage system that is shared between all nodes and is intended for the bulk of data for all your compute jobs.
There is no restriction in how much data you can put on that system, but there is a time limit.
Your data on the Lustre system is organized in so-called workspaces. A workspace is a directory that you create using the hpc-workspaces software. You will have to specify a name and a duration for each workspace.
On the default file system, a duration of up to 90 days can be specified. You can extend the duration up to three times in 90-days increments, for a total of 360 days. Please keep in mind your responsibility to not waste resources.
When a workspace expires, the data will be kept for another 21 days, then it will be deleted. We highly recommend setting up e-mail reminders, see below.
See also the hpc-workspaces User Guide.
To create a new workspace, use the ws_allocate command:
ws_allocate <workspace name> <duration in days>
The command will give you a prompt telling you the name of your new workspace directory.
Example:
$ ws_allocate demo 90
Info: creating workspace.
/lustre/scratch/data/jsteine1-demo
remaining extensions : 3
remaining time in days: 90
Double-check that you do not forget the duration, as otherwise the workspace will be allocated only for the default duration of 1 day.
The Lustre system has three different file systems. You can optionally select on which you want to create your workspace by using the -F option to ws_allocate. If you leave it out, it will go onto the scratch file system, which is suitable for general purpose use.
Example:
$ ws_allocate -F mlnvme demo2 90
Info: creating workspace.
/lustre/mlnvme/data/jsteine1-demo2
remaining extensions : 3
remaining time in days: 90
The Lustre system is already connected to the compute nodes with our fast Infiniband network. If you need a very high file I/O rate, you can make use of the MLNVME file system, which consists of SSDs.
| File system name | Technical specs: | Purpose |
|---|---|---|
scratch |
HDDs, 5 PB total | Default |
project |
HDDs, 5 PB total | Special purpose, only admins can create workspaces here |
mlnvme |
SSDs, 307 TB total | For jobs that require large amounts of file read/write operations |
You can make your workspaces available to other members of your group by using the -g and -G options, as described in the hpc-workspaces user guide under "cooperative usage". Marvin does support ACLs in the /lustre filesystems.
As described in the user guide, you can use ws_list to list your workspaces, ws_extend to extend, ws_release to release and so on. Note that the commands have man pages.
Note that when you extend a workspace, you also need to specify the same filesystem, i.e. the same -F parameter, as when you allocated it. If you left this out when allocating, you can leave it out when extending too.
The name that these commands expect for the workspace is the same name you specified when allocating it. So without the prepended user name and minus sign in the folder name the workspace resides in.
You can extend the duration up to three times, but please keep in mind your responsibility to not waste resources, and delete unneeded data.
When a workspace expires, the data are deleted after 21 days. If you lose a workspace, you may get in contact with us, and if you are quick enough we can restore it.
You can set up the workspace mechanism to send you e-mails a few days before your workspace expires.
We highly recommend setting e-mail reminders for all your workspaces, as it reduces the risk of data loss.
To set an e-mail reminder for a specific workspace, use the -r and -m options in the ws_allocate call.
Tip: you can make this even easier by creating a user configuration file in your home directory that contains your e-mail address, and optionally a default duration.
To do that, create an empty text file named .ws_user.conf in your home directory. Then set whatever parameters you want inside that script, see the user guide for all possibilities.
Example .ws_user.conf:
mail: demo.user@uni-bonn.de
reminder: 7
This will result in you getting an e-mail reminder 7 days before each workspace expires, assuming you did not specify something else when allocating it, of course.
Note that the user conf file needs to follow proper YAML syntax, for example there needs to be a space after the :.