Mass Storage System - Gyrfalcon
The Mass Storage System, known as Gyrfalcon, is the primary system for long-term storage of user data in the NREL’s high-performance computing (HPC) data center.
Gyrfalcon is designed to keep the most frequently used data quickly accessible and to economically store data that is accessed less often. This is achieved by employing high-performance disks for the freshest working data and by moving older data to economical tape storage with the help of software algorithms and the system configuration.
This mass storage system is available to all HPC system users and projects. Users without an account may also request access.
- Both the amount of data and the number of files will be managed with quotas, and quotas are enforced by the group of the files.
- Users with accounts on any NREL HPC system will be provided with accounts on this system for personal storage. Every system user receives a default 1 terabyte, 1 million file quota automatically when their account is created. To request additional space, contact us.
- Projects that receive an allocation on HPC systems will be provided with space on this system, with the project quota determined as part of the allocation process. Allocations receive a default 2 million file quota when the allocation is created. To request additional space or file quota, contact us.
- Gyrfalcon is not backed up. Two copies of all data are stored on different media; however, if a user deletes their files, there is no backup from which to restore them.
The Gyrfalcon system includes a 1 petabyte disk storage system as well as an Oracle StorageTek robotic tape library with seven high-performance T10000C tape drives and seven T10000D tape drives. It uses Oracle's QFS file system and SAM archiving software, which allows users to simply copy data to and from the file system without knowing or worrying about which tier the data is stored on. It has a capacity of over 3 Petabytes of user data, with an architecture that allows the capacity to be easily expanded at relatively low cost.
Frequently Asked Questions
The disk system capacity is 756 TB and the tape capacity of this system is 5.2 PB. By policy, two copies of each file will be stored on different media, with copies being handled automatically by the archive software. This provides over 2.5 PB of long term data storage.
The /mss file system is mounted on login nodes at "/mss". Each user has a user directory under /mss/users/$USER.
Because /mss is an regular file system, ordinary Linux commands such as cp, mv and rsync may be used to transfer data from /home or /scratch file systems to the /mss file system.
At the command line of the system login nodes, enter one of the following commands to copy files to mass storage. Please note that these commands may take several minutes to several hours to complete.
Option 1: The first command will create a list of files in a directory. The second command (tar) will use that list to gather all the files in the directory into a single file which resides in the MSS directory indicated. The files in the original directory are left unchanged.
$ lfs find directory > directory.txt
$ tar –czf /mss/<MSS directory>/directory.tgz -T directory.txt
Option 2: The first command (tar) will gather all files found in a directory into a single file. The second command will copy the resulting tar file onto the mass storage system. Alternatively, replace "cp" with "mv" to move the resulting tar file from the original directory to the MSS directory. As with option 1, the files in the original directory are left unchanged.
$ tar –czf directory.tgz directory
$ cp directory.tgz /mss/<MSS directory>
Option 3: The rsync command compares one directory to another and makes the destination directory equal to the source directory.
$ rsync –av directory /mss/<MSS directory>
Option 4: The simple Linux cp command can be used to copy a file from one directory to another directory. This command is best used for small numbers of files.
$ cp filename /mss/<MSS directory>
Users may access their /mss directories via a server - mss1.hpc.nrel.gov.