2.4) ARC2 Hardware: Storage
Home directory: Currently, each user on ARC2 has a home directory of 5GB. This home directory is backed up regularly and should be used for high value files, such as source code, and possibly as a staging area for data transfers or input data. This home directory is accessible via the standard environment variable $HOME, and in longhand /home/<some_server>/<user_name>, where <some_server> will change from user to user, and <user_name> will be your user name on ARC2. This directory is automatically created by the system when an account is created.
/nobackup: There is additional storage in the form of a large parallel Lustre filesystem. This filesystem consists of ~170TB of usable high speed temporary disk space. It is configured to deliver access to files, i.e. Input/Output (I/O) operations, at a rate of up to 4GB/s via the InfiniBand network. The main purpose of this file system is to provide a spooling area for running processes/applications. Providing a high rate of I/O can greatly improve the performance of applications, in particular those that are limited by I/O, or so called I/O bound problems. I/O bound problems are those whose progress is limited by the speed of the I/O subsystem. For instance, a task that processes data from disk, such as counting the number of lines in a file is likely to be I/O bound. This filesystem is mounted on /nobackup, and as the name implies it is not backed up! There are no quotas imposed on /nobackup, but files unused for 90 days will be “expired” or deleted by the system, after a two and one week warning. It important to stress that /nobackup should not be used for medium or long term storage, and never for the sole copy of high value files, such as source code. To make use of /nobackup, simply create a directory.
mkdir /nobackup/<user_name>
/scratch: Each compute node has a local disk attached and mounted on /scratch. These are typically quite small, (~400Gb) and shared between all jobs running on a node. They are the best option when running processes that require a high rate of small I/O operations. It is important to note that this is temporary storage and is only accessible to the node on which the disk is mounted. Files written should be deleted/removed at the end of the process.