Input data
==========

Each cluster user can use the ``~`` as a storage of the input and output
data for experiments. Please note, that the ``/home`` partition is an NFS
share on compute nodes. Because of this fact each time an experiment is run all
input data must be copied again from the login node.

Alternatively, most of the compute nodes have a local partition (or two),
available for each user. Any data can be copied to this directories (using
an interactive session). This approach tries to minimize the
network load and speed up experiments.

.. warning::
   Please remember that the ``/home`` directory is acessible
   through the Faculty's network. Running a job that requires a large input dataset twice,
   will cause downloading this dataset twice. When using even moderatly big input data, please
   consider using ``/storage_type_N`` filesystems.

/storage_type_N directories
----------------------------

This directories (``N = 1, 2, ...``) are disk partitions local to a compute node.
They are available to all users and no quota is enforced.
This filesystem **is not** synced with any other filesystem and has multiple uses, for example:

+ is freely available during interactive sessions,
+ can fit large input data, virtual environments, etc. created during 
  interactive sessions,
+ is suitable for any kind of results, if they are larger than default limits imposed
  on the ``/home`` filesystem (but not only),
+ any data can be copied outside the cluster using ``scp``, ``rsync``, etc.;
  there is no automatic system configured to transfer or delete this data.
  Please use an interactive session to manage files.

But please be considerate of other users, do not use too much space and delete
unnecessary data. Please use ``--nodelist`` to select a specific node.

.. note::
   As this filesystem is available to all users, a common sense and
   awareness of others should be exhibited by all users of the cluster.
   
   * One should delete redundant files.
   * There is probably no need for several copies of the same data.