Where can I read/write data to/from?
On Atlas, we have a large number of storage devices and it is not easy to keep track of all of them. Therefore, here is a list of possible storage paths along with some information about each.
For each of these we will use two variables. $USER
refers to the
username within Atlas and $NODE
refers to the name of a machine,
i.e. the generic path /atlas/user/$NODE/$USER
could mean
/atlas/user/a1234/albert.einstein
, i.e. a specific directory named
albert.einstein
on machine a1234
.
Please note, that most parent directories of these file systems are “virtual” ones, this may mean that these directories appear empty until one enters a sub directory. In other words, you may need to know the exact name of the directory to enter beforehand!
writable locations | read-only locations |
---|---|
/work/$USER |
/home/$USER |
/atlas/user/$NODE/$USER |
/atlas/ldr |
/cvmfs |
|
/opt/ |
/home/$USER
- Availability
- head nodes only
- Physical storage
- Files formerly stored on the HSM
- Distributed on three 20Gbit/s connected servers
- Usage
- read-only data store
- data should soon be archived by project or copied to /work
- Expected life-time
- at most until 2023
- Access via web browser
https://www.atlas.aei.uni-hannover.de/home/$USER
- for details please refer to User WWW directories
/work/$USER
- Availability
- head and compute nodes
- Physical storage
- Around 30 servers
- Files stored on dedicated 24 disk storage boxes
- Total bandwidth 20Gb/s
- about 84 TByte disk storage
- Usage
- scratch space
- input and final output of Condor work flows
- Expected life-time
- TBD: automatic snapshots, files can be retrieved by users after accidental deletion
- Access via web browser
https://www.atlas.aei.uni-hannover.de/work/$USER
- for details please refer to User WWW directories
- Details
- More details can be found the dedicated page for
/work
/atlas/user/$NODE/$USER
- Availability
- across cluster
- access to head nodes’
/local/user/$USER
directory is not possible - Alias of
/local/user/$USER
on machine$NODE
- Physical storage
- compute node (single disk, ~1 Gbit/s, no redundancy, ~1 TByte)
- head node (12-16 disks, ~20 Gbit/s, some redundancy, ~50 TByte)
- remote access via NFS, locally via bind-mount
- Usage
- local scratch space
- intermediate files created during condor work flows (compute nodes)
- condor log files, git clones for temporary builds (head nodes)
- no backup is made for these
/atlas/ldr
- Availability
- across cluster
- Physical storage
- 37 servers
- 12 disk drive raidz2
- optimized for reading
- 20 GBit/s per machine
- Usage
- storage of detector data (mostly LIGO/Virgo)
/opt
- Availability
- across cluster
- Physical storage
- single, read-optimized server
- 20 GBit/s per machine
- Usage
- central storage area for application software
- Mathematics, Matlab
- commercial compiler
/cvmfs
- Availability
- across cluster
- Physical storage:
- virtual file system
- provided by local caching and multiple proxies
- Usage
/cvmfs/oasis.opensciencegrid.org/
for example contains software from multiple experiments