|
In order to keep filesystems from filling up and disrupting work, we use filesystem quotas to limit usage in certain areas by user or group. This also helps us observe growth in disk usage over time so we can plan future expansion.
By default, the filesystem quotas are as follows:
filesystem | quota (maximum total data size allowed) | file limit (maximum number of files allowed) |
---|---|---|
| 100 GiB per user | none |
| varies by lab/group | none |
| varies by lab/group | none |
| 10 TiB per user | 1,000,000 files or directories |
| 15 TiB per user* | none |
* scratch_gpu
is only available for labs whose PI has a primary or secondary appointment in a pre-clinical HMS department.
...
For data on /n/scratch3
, you need to use the scratch3_quota.sh
command:
Code Block | ||
---|---|---|
actionscript3 | actionscript3 | $ /n/cluster/bin/scratch3_quota.sh Directory: /n/scratch3/users/m/mfk8 Space used: 0TiB used of 10TiB Files/directories: 1256 of 1000000 |
...
Note: It is against RC policy to artificially refresh last access time of any file located under /n/scratch3.
For more information on scratch3, please refer to the dedicated scratch3 wiki page.
scratch_gpu quotas
Quota utilization for /n/scratch_gpu
is reported by the command quota if used from a node where the /n/scratch_gpu
filesystem is available (login, transfer or compute-g nodes)
...
Another way to check usage is to total the size of files in a directory using the du
command. For example, you might want to see how much space your sub-directory in your group's shared directory is consuming:
To check the size of a directory (e.g.
/n/groups/smith/mydirectory
):Run the command:
du --apparent-size -hs /n/groups/smith/mydirectory
The output returned is the total size.
Note that
du
can take quite some time for directories containing large numbers (tens of thousands or more) of files, because it must check the size of every file to compute the total. In general, it is better to usequota
to find usage information, when possible, or at least to rundu
on sub-directories instead of top-level directories.
The
--apparent-size
option is required to find files' actual sizes. Without this option, the reported size will include data protection overhead (redundant copies of data on the O2 file server, which protects against hard drive failures).Please do not run
du
from a login node. Long running and computationally intensive processes will be killed on login nodes. To ensure that your command for checking directory usage is not interrupted, please rundu
from a compute node instead. You can use thesrun --pty
command to start an interactive job, and then rundu
once you have been allocated resources. More information on running SLURM jobs can be found on the Using Slurm Basic wiki page.
When you are over quota
How to tell you are over quota
...
Note that the quota
command results are only updated hourly. If you were writing files very rapidly, the quota
command might not show a completely full quota. Also, deleting files won't immediately change the results from that command. If you delete 5 GiB of files, you should be able to write 5 GiB of new files in that location immediately, even if quota
hasn't caught up yet.
You can delete a whole directory with a command like rm -rf dir
. Please be careful when using a command like this: you could delete all of your files!
...
For additional information on HMS RC storage offerings, and where O2 cluster users can store data, please reference the following pages: