- Introduction to HPC clusters
- UNIX Introduction
- Condo 2017
- Classroom HPC Cluster
- File Transfers
- Cloud Back-up with Rclone
- Globus Connect
- Sample Job Scripts
- Using DDT Parallel Debugger, MAP profiler and Performance Reports
- Using Matlab Parallel Server
- Using ANSYS RSM
- Support & Contacts
- Systems & Equipment
- FAQ: Frequently Asked Questions
- Contact Us
- Cluster Access Request
Currently each user has 5 GB in home directory space (/home/<username>). This is for configuration and login files. It will be quite a bit slower than other disk resources, so it should not be used for high volume access.
Long-term Group storage
Each group has a directory shared by all members of the group (and only that group). These directories are links under the directory /work. Issue "ls /work/" to see all group directories. Unless you're in the LAS group, "cdw" command will cd to /work/<your_group_working_directory>/<NetID> . To find your group, issue the "groups | grep nova" command. Normally groups will have names its-hpc-nova-<NetID>, where <NetID> is NetID of the PI for this group.
Group working directories are available on all nodes, and any user in the group can create files there using the group's quota. Group quota for this space is based on the shares for the group and can be seen at login. The usage and quota can also be found in the file /work/<your_group_working_directory>/group_storage_usage which is updated once every hour.
The group working directories are backed up daily using ZFS snapshots. The snapshots contain the previous version of a file, or in the case of a deleted file, the last version of it. Use of snapshots allows to do incremental backups based on the these snapshots, which is needed because there are several millions of files on each server. The snapshots are also for filesystem consistency, as the file can be reverted to the previous version by the filesystem admin. The snapshots are copied to a backup storage and are deleted from the primary storage after 5 days. The snapshots on the primary storage use group allocation. That means that removing files frees up space only after 5 days.
The ZFS on the storage servers uses automated compression, so users can actually store more data in the same amount of physical space. The compression ratio depends on the data. Autocompression not only saves space, but also improves performance. However any file system will perform worse when it's close to being full. For this reason the group's PI will receive an email when group's data usage is over 70% quota. It's a good practice to keep file system less than 70% full.
Temporary local storage
One-node jobs can use the storage on the disk drive on each of the compute nodes by reading and writing to $TMPDIR (about 1.5 TB on most of the compute nodes, 11TB on the high memory node). This is temporary storage that can be used only during the execution of your program. Only processes executing on a node have access to this disk drive. You must ensure that multiple processes executing on the same compute node don't accidentally access files in $TMPDIR meant for other processes. If running MPI program on a signle node, one way to accomplish this is by including MPI process ranks in temporary filenames.
Temporary parallel storage
Multi-node jobs will build a parallel file system on-demand which utilizes the local storage on each compute node assigned to the job. This provides a large fast-access scratch space for temporary files that resides alongside the local storage available in $TMPDIR. This filesystem will be mounted at /mnt/beeond during the execution of the multi-node job. It is recommended to access this location using the environment variable $PTMPDIR.
To use this parallel storage the following workflow should be used. These steps may be taken interactively (when salloc'd to a compute node) or in batch-mode. In batch mode the copy commands below should be added to the job script.
Copy calculation input to the parallel filesystem.
e.g., cp /work/<your_group_working_directory>/<user>/<input files> $PTMPDIR where <input files> contains the folders/files to be used in your calculation (to copy the whole folder use "-r" option).
Run your code, getting input from files located in $PTMPDIR and writing output to $PTMPDIR.
Copy final results to storage location
e.g., cp $PTMPDIR/<final results> /work/<your_group_working_directory>/<user>/<final results>
Note that this parallel filesystem will disappear at the conclusion of your job. Any data which is not copied out of $PTMPDIR cannot be recovered after your job has finished.
Myfiles are mounted on the Nova cluster. To access your directory, ssh to novadtn (you won't be prompted for the password). When on novadtn, issue:
and enter your ISU password when prompted. To access your MyFiles directory, use /myfiles/Users/<your_user_name> or /myfiles/<your_dept>/users/<your_user_name> (depending on the department, the path may differ from the one above). Additional information on MyFiles can be found on the Information Technology website.
When transfering files to or from nova storage please use the Data Transfer Node, novadtn.its.iastate.edu. All of the storage options are availbile there and it will help keep the head node from becoming overburdened.