Technical Bulletin #538: VAST File System
Update Text below in purple was updated on 8/6/20.
Update Text below in bold green is a further addendum on 8/6/20.
Livermore Computing has deployed a new file system, called VAST, on LC CZ systems. VAST is a fundamentally new file system architecture that incorporates many innovative technologies and features, which are detailed below.
VAST uses flash storage exclusively in place of more traditional file systems which use spinning disks. Its initial capacity will be 5.2 PB, however usable capacity is expected to actually INCREASE over time due to VAST’s novel similarity compression algorithms. All users with accounts on CZ systems will have a directory under /p/vast1. VAST uses the NFS protocol and should behave like any other NFS file system e.g. your home directory or /usr/workspace. VAST can also be used in place of Lustre for parallel I/O.
The mount point for this file system on CZ clusters is /p/vast1. Currently the VAST file system only implements directory quotas. LC has set User and Group directory quotas at 20TB. In the future VAST will offer the option of LC implementing user quotas. When user quotas are implemented, LC will evaluate on a case by case basis any user who has an aggregate amount of data > 20TB stored across the file system. VAST has demonstrated very good performance and stability in limited testing on Corona. Now that VAST is mounted on the CZ production clusters we encourage users to try it out for various use cases and send your feedback to email@example.com. We would be very interested in performance comparisons between VAST and Lustre. VAST also provides an S3 object storage interface. If you have use cases for an object store and want to try out VAST, please let us know!
More information about VAST file systems can be found on their whitepaper here: https://vastdata.com/whitepaper/