Storage¶

Home directory¶

Your home directory is accessible et $HOME. For instance:

[alice@ip09 ~]$ echo $HOME
/home/alice

This is the right location for configuration files, your code, and software that you install. Due to limited capacity and performance, it is not the right location for research data and you should not use it to read and write such data when running jobs.

Research data¶

Research data is stored on a 190T server accessible at /net/nfs-iq/data. Use this locaton to read and write data when running jobs; it offers better IO (input/output) performance than your home directory.

To ensure safety in case of a drive failure, research data is redundant, that is to say that two copies are stored. (The total physical capacity of the server is 384T, roughly 2 × 190T.) Data is also backed up daily, and archived on tape weekly.

Individual data¶

Each user has their own individual space for research data, accessible at /net/nfs-iq/data/$USER. For instance:

[alice@ip09 ~]$ echo $USER
alice
[alice@ip09 ~]$ cd /net/nfs-iq/data/alice

Group data¶

Research groups have a shared space accessible at /net/nfs-iq/data/def-$PROF, where $PROF is the group leader’s username. (This form is identical to that of project storage on Alliance clusters.) For instance:

[alice@ip09 ~]$ echo $USER
alice
[alice@ip09 ~]$ cd /net/nfs-iq/data/def-alice

Local storage on compute nodes¶

When the scheduler starts a job, a temporary directory is created on each allocated compute node. The SLURM_TMPDIR environment variable gives the directory’s path. When the job ends, files in the temporary directory are deleted.

Since $SLURM_TMPDIR is on a local disk, input/output operations are faster than on network storage (such as /net/nfs-iq/data). If a job performs many read/write operations, it will probably run faster using $SLURM_TMPDIR rather than network storage. You must however make sure to copy your data to network storage before your job ends.

MP2 cluster interconnection¶

Research data storage in /net/nfs-iq/data is accessible from the MP2 login and compute nodes. MP2’s /project and /scratch storage are not accessible from the IQ HPC Platform.