Linux Cluster Disk Guide
The most important things to learn from this page:
- /home and
/share
partitions are backed up.
- /data and
/scratch
are not.
- The disk quota for temporary and guest accounts is 10GB.
This is a guide to issues with disk storage on the on the
Linux cluster.
- To find out how disks on one system can be accessed from another, see the automount page.
- To understand different partition names (e.g., why are there
/share
and /scratch
directories), see the disk sharing page.
- To learn about the server for temporary and guest accounts, see the student file server page.
How much disk space do I have?
To find out how much disk space you have available, use the
df
command. You'll probably always want to use the
-h
option, so the sizes appear in human-readable form:
df -h
You'll almost certainly see disks in the list that are mounted via
automount. If you find the automounted disks to be distracting, add
-l
to the command:
df -hl
Bear in mind that you
don't want to use the
-l
option if your home directory is not on the machine to which you've logged in. (As of Jan-2017, this mainly applies to ATLAS users logged onto
xenia
.)
Here's the result of executing
df -hl
on the machine
tanya on 28-Jan-2017:
Filesystem Size Used Avail Use% Mounted on
/dev/mapper/VG-root 20G 9.2G 9.5G 50% /
tmpfs 494M 760K 493M 1% /dev/shm
/dev/md0 461M 98M 340M 23% /boot
/dev/mapper/VG-home 50G 16G 32G 33% /home
/dev/mapper/VG-data 222G 6.1G 205G 3% /data
If we ignore partitions that relate to the operating system, we're left with two key user-accessible filesystems:
/home
and
/data
. (Many systems have other key partitions, such as
/share
and
/scratch
.)
It's not much
Your first reaction may be: "There's not much disk space for my home directory, and I have to share that space with other people in the collaboration. Why, my watch has more storage than that!"
You're right. It's intended that the
/home
be used for "source" files (program code, scientific papers, plots, etc.);
/data
or
/scratch
should be used for large and re-creatable files (compiled binaries, data summaries, temporary work files, etc.). We have to ask you to use judgement and discipline, and to be aware that you're sharing space with your fellow scientists.
If you're just skimming this page, stop and read this
The reason why
/home
is small and
/data
is big is that the
/home
partition is backed up;
/data
is not. In fact, it goes one step further: the
/data
partition is always considered expendable for any type of system maintenance activity. If a system is being repaired, upgraded, or restored, the
/data
partition may be erased.
There's more about this in the section on backups below.
What do I do if I need more disk space?
First, look to
/data
partitions on other systems in your working group. The
/data
partitions on all the systems that belong to a group are intended to be a shared resource; if you don't have enough space on
/nevis/yourmachine/data
,
cd /nevis/othermachine/data
in your group and see how much free space it has.
I strongly advise you to exercise common courtesy as you're scrounging for disk space. If I found someone had used a big chunk of my server's
/data
partition without asking, I might be annoyed.
If you still don't have enough disk space on all your group's machines to satisfy your needs, you may have to request more disks be added to the existing systems (or buy a new box).
Backups
The Nevis Linux cluster sytems are backed up periodically onto
backup.nevis.columbia.edu, the Nevis backup server.
For speed, we don't copy every file from every system; we use a program called
backintime
to copy only those files that have changed over time. The
backintime
program functions similarly to Apple's
Time Machine
: Periodically, only those files that have changed since the last
snapshot
are copied to the backup server.
We don't back up every file on every system on the cluster. The practice is: the
/home
partition and
/share
partitions are backed up;
/data
is not. There is a web page that contains the
list
of which partitions are backed up.
The frequency with which we take backintime snapshots varies between systems and disk partitions. For example,
/home
and
/share
partitions have snapshots made every four hours; the cluster library directory
/usr/nevis
(whose contents change rarely) has a snapshot made once a day. Old snapshots are kept for a time, again based on how much space is available on the backup server; for example, as of Feb-2025 old files from
/home
partitions are kept for years; mail files (which can grow very large depending on user preferences) are only kept for only a few days.
- Why are
/home
partitions so small?
The answer is backup. There are roughly 100 systems on the Linux cluster, and we back them all up. Some of the backintime jobs take hours to run. Even if we had more disk space, as a practical matter we can't have a regular backup that takes more than 4 hours to run.
We therefore have to ask users to segregate their files into key files that will be backed up, and re-creatable files that won't. The relative sizes of /home
versus /data
partitions help enforce this segregation.
- Why don't you back up
/data
partitions?
We have vastly more disk storage on the Nevis cluster than we can hope to back up on any system that we can afford. As of Jan-2025 we have over 500TB of storage assigned to /data
partitions on different systems.
- I've got files in a
/data
partition that would be a pain to re-create. How can I back them up myself?
The simplest thing to do is to make copies on other /data
partitions in your workgroup's cluster. After all, that's all a backup is: a second copy of your files.
- Why don't we make backups more frequently, or keep files for longer?
We're doing what we can with the resources we have available. We don't have the disk space on our backup server for a year's worth of mail backups, for example.
- I've got a brilliant idea! I'll put a link in my home directory to a directory on a
/data
disk. That way, when my home directory is backed up --
Stop right there. It won't work. The backup procedure does not follow links. We've had at least one student who lost critical files because they tried this trick. You can't magically increase your available backed-up disk space in this way.
- I've got critical files that that are not in my /home directory, or that I want backed up even more often. What can I do?
Supplement our backups with your own. You can run the program backintime-qt
yourself, saving files from any source to any destination for which you have read/write permission. You may also want to look at the documentation for command-line backintime
and details of the backintime configuration file
.
Take care! Think carefully about what you include in your snapshots and how long you want to keep them. Without limits, backintime will fill up an entire drive before it deletes excess snapshots. If other users share the same disk space, this may irritate them if you're doing four-times-an-hour backups of your frequently-refreshed 10GB files.
Long-term data storage
For the purposes of this section, "long-term" means more than six months or so.
By the above definition, there is no long-term data storage at Nevis. As noted above:
- we back up
/home
directories, but keep we don't keep snapshots indefinitely;
-
/data
directories are not backed up at all;
- RAID arrays can and do fail. (This section is being written on 25-Apr-2006; on that day, we lost the contents of a RAID5 array.)
If you need long-term storage for any of your files, I suggest you consider the facilities at BNL, FNAL, or CERN.