Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...

Table of Contents

Overview of  /n/scratch_gpu filesystem

The /n/scratch_gpu is a storage filesystem on O2 dedicated for temporary GPU computing data. There is no maximum retention period for data that is being processed, however any data idle for more than 30 days will be automatically deleted. 

Warning
  • The scratch_gpu space is NOT BACKED UP!
  • Files are AUTOMATICALLY DELETED 30 days after they are last accessed.
  • There are NO SNAPSHOTS for data stored under /n/scratch_gpu

Data that is deleted on purpose, by accident, or automatically after being idle for 30 days is GONE and cannot be recovered. Using this temporary data retention model allow us to provide this cost-effective but high-end storage system for the HMS Community. 

Note

Artificially modifying last file access times (timestamp) is against our policy and may result in losing your access to /n/scratch_gpu

Note

/n/scratch_gpu will only be available on login nodes, transfer nodes, and GPU nodes.

Eligiblity

At this time, the  /n/scratch_gpu space is only available for Labs whose PI has a primary or secondary appointment in an HMS pre-clinical department.

Quota limit for /n/scratch_gpu

Each User has a default quota of 15 TiB (~15.3 TB) for files and folders stored under /n/scratch_gpu

Once utilization reaches the default 15TiB quota, the user will be notified via email and will have a grace period of one week to delete data. After this week the User will not longer be able to create any new data until the usage is reduced below the 15 TiB default quota. If the data exceeds 16TiB the User will immediately be unable to create any new data until the usage is reduced below the 15 TiB default quota, regardless of when the default 15TiB quota was exceeded.

You will be able to see your /n/scratch_gpu usage with the command quota; however quota information about /n/scratch_gpu will only be available on O2 nodes where the filesystem is available. 

Create a User /n/scratch_gpu  Directory

To create your own /n/scratch_gpu directory on O2, please run the following command from a login node (not an interactive job):

/n/cluster/bin/scratch_gpu_create.sh

How to use the /n/scratch_gpu filesystem

On scratch_gpu, user directories are found under:

  • /n/scratch_gpu/users/<first_HMSaccount_id_char>/<HMSaccount>

Where "<first_HMSaccount_id_char>" is the first character of your HMS account id and  "<HMSaccount>" is your HMS account ID.

For example, the scratch_gpu directory for a user with the username "foo123" would be located at:

  • /n/scratch_gpu/users/f/foo123

...

Each User must create her/his scratch_gpu directory using the script /n/cluster/bin/scratch_gpu_create.sh as described in the paragraph above.

...


Research Computing is no longer providing the /n/gpu_scratch filesystem. Please use /n/scratch3 instead.