File storage and I/O
There are 5 main user file spaces available on Genepool/Phoebe
Your home directory on Genepool mounted across all NERSC systems. You should refer to this home directory as $HOME where ever possible. If you had an old home directory at JGI, on /house or on the netapps we recommend you refer to this old home directory as $OLD_HOME. You should not change the environment variable $HOME. For more details on quotas, dotfile initialization and back ups, please see our global homes page.
The house file system is mounted on Genepool as well as JGI systems. Users can access data from house as usual. Note that house is currently highly utilized and often is more than 90% utilized. Please backup data to HPSS and delete data you don't need. Consider using the projectb file system described below. Genepool is the only NERSC computational resource that has access to /house. If you want to use hopper or carver, please migrate your data to projectb.
projectb is a new 2.7PB GPFS based file system for JGI projects. There are two portions of the projectb filesystem: projectb/projectdirs and projectb/scratch. The projectb filesystem is available on genepool, hopper and carver. The projectb filesystem is only mounted on NERSC computer systems.
|projectb Projects||projectb Scratch|
|Quota||5TB default||20TB, 5M inodes by default; 40TB upon request|
|Backups||Daily, only for projectdirs with quota <= 5TB||Not backed up|
|File Purging||Files are not automatically purged||Files not accessed for 90 days are automatically deleted|
The intention of the projectb "Project" space is to put data that is needed by multiple people collaborating on a project to allow for easy access for data sharing. If you would like a project directory, please use the Project Directory Request Form.
projectb "Scratch" space is intended for staging and running JGI calculations on the NERSC systems, including genepool, hopper, and carver. On genepool, the projectb scratch space is the recommended filesystem for performing file IO during all your calculations. If you have access to the genepool resource, you should have space on projectb scratch. If you don't, please file a ticket at http://help.nersc.gov.
Each user has a "scratch" directory. Scratch directories are NOT backed up and file are purged if they have not been accessed for 90 days. Access your scratch directory with the environment variable "$SCRATCH" for example:
Scratch environment variables:
|Environment Variable||Value||NERSC Systems|
|$SCRATCH||Best-connected file system||All NERSC computational systems|
|$BSCRATCH||/global/projectb/scratch/<username>||genepool, hopper, carver|
|$GSCRATCH||/global/scratch/sd/<username>||All NERSC computational systems|
$GSCRATCH points to your Global scratch space, and $BSCRATCH points to your projectb scratch space if you have a BSCRATCH allocation. $SCRATCH will always point to the best-connected scratch space available for the NERSC machine you are accessing. For example, on genepool $SCRATCH will point to $BSCRATCH, whereas on carver $SCRATCH will point to $GSCRATCH.
The intention of scratch space is for staging, running, and completing your calculations on NERSC systems. Thus these filesystems are designed to allow wide-scale file reading and writing from many compute nodes. The scratch filesystems are not intended for long-term file storage or archival, and thus data is not backed-up, and files not accessed for 90 days will be automatically purged.
The /jgi/tools area has been brought over to Genepool as is. Users can access software as usual.
Other file systems
Other file systems will also be mounted on Genepool, though they may be lightly used by JGI users initially.
/usr/common - is a file system where NERSC staff build software for user applications. We expect to build out this area for the JGI in the future.
global scratch - is a GPFS based file system that is accessible on almost all of NERSC's other compute systems. Write performance to global scratch on Genepool will be slow. However, using this file system could be an option for users needing to share data with other NERSC systems.
global project - is a GPFS based file system that is accessible on almost all of NERSC's other compute systems. Write performance to global scratch on Genepool will be slow. However, using this file system could be an option for users needing to share data with other NERSC systems.