Tungsten File System Overview

User Information Home
Compute Resources
Software
Data
Security
Allocations
Consulting
Training
Strategic Applications Program

Tungsten login nodes are served by an NFS file system. I/O from batch runs should be directed to one of the LUSTRE-mounted directories (see table below). Currently the LUSTRE file system on Tungsten is served by a total 104 Object Storage Devices (OSTs) split over two mount points (/cfs/scratch and /cfs/projects/widescratch).

Tungsten File System Information

Name	Location	Environment Variables	Size	Policy
				Quota	Purge	Backup
Home Directory	/u/ncsa/$USER	$HOME	5GB per user	5 GB	N/A	Daily
Scratch (LUSTRE)	/cfs/scratch/users/$USER    also linked as: $HOME/scratch-global	$TG_CLUSTER_ SCRATCH	59TB	None	7 days	None
Wide Scratch (LUSTRE w/ striping)	/cfs/projects/widescratch/$USER	None defined	49TB	None	7 days	None

Usage

The $HOME/scratch-global directory is linked to a LUSTRE mounted directory. Directing I/O from batch jobs to this directory is a good general practice. For more complete usage information refer to User Documentation.

Performance Notes

The default scratch directory on Tungsten is mounted on a LUSTRE file system that does not stripe files across OSTs, which provides good general usage performance especially if one or more files per process are accessed. However, applications accessing large files from many processes will not see performance gains from this setup. In this case, using the striping capability of LUSTRE may improve performance. See /usr/news/Lustre_Striping on Tungsten for complete details.