Data and disk filesystems

Total disk space of cluster Snehurka is based on data storages with higher reliability (RAID5, redundant power supplies, etc.) and is divided into two filesystems:

• /usr/users - regularly backed up and archived filesystem with home directories of all users in the form /usr/users/login - this directories are intended for your worth data (scripts, initial data, configuration files, and other important files whose acquisition cost - human or computer - is high)
• /usr/work - larger filesystem with working directories of all users in the form /usr/work/login - this directories are assigned for data from your computations, debug files, temporary tools, etc. - everything what can be "easily" recovered

Because disk space for filesystem is shared by all users. For its optimal utilization we rely on users considerateness. To enable everyone to use temporarily rather big amount of disk space without administrative obstructions, please,

• delete all data which you will not need any more,
• compress data, which you need to keep untouched on cluster for some time (results of PhD. thesis before you defense it, etc.)
• be conscious, that home directory are backed up and archived (for periods of weeks or month) and therefore require other sources

For smooth running of filesystems and optimization of space sharing among users soft quota system is introduced (see end of page how soft quota system works).

Some useful hints:

To find out the size of free space on filesystems

df -h /usr/users /usr/work
beegfs_nodev 35T 28T 6.6T 81% /srv/beegfs/cluster
beegfs_nodev 35T 28T 6.6T 81% /srv/beegfs/cluster

Find out data size

r0:/usr/work/login/fenics> du -s -h results/
196M	results/

Compress directory

Useful for reduction of size of data on disk (valuable especially for ascii files):

r0:/usr/work/login/fenics> tar -cvzf results.tar.gz results
r0:/usr/work/login/fenics> du -s -h results.tar.gz
58M     results.tar.gz

or using more effective but slower compress method

r0:/usr/work/login/fenics> tar -cvJf results.tar.xz results
r0:/usr/work/login/fenics> du -s -h results.tar.xz
11M      results.tar.xz

Remove directory with subdirectories

(be careful if you use asterisk in command line)

r0:~/Work/fenics> rm -r results

Uncompress data

tar -xvzf results.tar.gz

or

tar -xvJf results.tar.xz

List content of directory sorted according to size of items

r0:~/Work> du -s -k * | sort -n
8	guess_id
16	Globals.mk
5004	bender
35532	bender_pressure
100108	bingham_ale
113976	git_repo
117500	implicit
649780	fenics

How soft quota system works

• Every user has set up soft limit for his data. Default values are
  10GB per login on directory /usr/users/login
  80GB per login on directory /usr/work/login
• If amount of your data is less than soft limit, nothing happens
• If your data exceeds soft limit for less then last 10 consecutive days, only the system administrator gets the information about it
• If your data exceeds soft limit for last 10-24 consecutive days, you get informative email about it (system administrator too)
• If your data (in /usr/users) exceeds soft limit for last 25-31 consecutive days, system will not backup your data and you get informative email about it (system administrator too)
• If your data (in any filesystem) exceeds soft limit for more than 31 consecutive days, your account will be disabled including mail delivering and you will have to contact system administrator to fix the problem, system administrator gets the information about it and waits

These rules mean:

• you computing jobs will not stop when your data exceeds soft quota limit
• you can use temporarily (for a few days) more space than soft quota guarantees (system administrator is informed)
• you will get info that you permanently run over quota limit (see above)
• you have enough time to reduce the amount of data under soft quota limit (or ask us to increase your soft quota limit) to come back to normal situation
• in the case your behaviour could influence other users or system performance, system administrator will take some measures (depends on seriousness of situation from sending email to immediate stoping your jobs)