Start Here

Storage for your data works a little bit different on a supercomputer compared to your desktop. A high-level introduction can be found below. This page also covers best practices around how to handle your data, this is the 3-2-1 policy. Each HYAK cluster has its own storage, refer to the klone or mox storage page for more cluster-specific in-depth details.

What is storage for a supercomputer?

Storage on every HYAK cluster is physically separate. It is best practice on every supercomputer that storage live as its own infrastructure to be high-performance and able to handle the bandwidth I/O and read/write operations required by so many compute nodes attached to it. These are typically parallel file systems (e.g., GPFS, Lustre, BeeGFS).

Storage systems are mounted (i.e., accessible) from every compute node of the cluster. Each HYAK cluster (e.g., klone, mox) has its own separate parallel file system. The storage attached to each HYAK cluster has its own policies, hierachy, etc. Please refer to their respective pages for more information.

warning

Cluster storage is not backed up!

While our storage systems have a track record of stability, it is important to note that STORAGE IS NOT BACKED UP by default. It is the responsibility of the user that in the event of an incident you have a place and plan to restore their data. We provide a complementary archive service that is appropriate for this and other solutions exist.

3-2-1 Policy

Your data is precious, in some cases completely irreplacable. The research computing team encourages the use of the widely accepted 3-2-1 backup strategy.

tip

3-2-1 is not a HYAK thing, it's a general IT best practice [Backblaze] [Acronis] [Networkworld].

The 3-2-1 backup policy suggests 3 copies of your data on 2 different types of storage media of which 1 copy is off-site. If you use both gscratch and LOLO then you are already adhering to this best practice, which is why it was designed this way. One copy resides in gscratch on our parallel file system and if you archive your data to LOLO, two additional copies are created (it does automatic duplication with one copy on UW-Seattle campus and another copy in eastern Washington). gscratch consists of spinning disk hard drives as a storage medium while LOLO is a tape-based storage medium. LOLO does one of its automatic duplication copies to a geographically remote data center in eastern Washington.

caution

You have to copy your data to LOLO to be 3-2-1 compliant or to use your own archive solution, it does not happen automatically.