Reliable Storage for CIBR Clusters

The home folders on all of the CIBR clusters are configured as scratch drives. They are all RAID arrays, and thus have some protection/redundancy, but they are not backed up in any way. That is, there is a significant risk of a failure causing complete data loss. Historically we have encountered an unrecoverable failure of this sort 3 times in ~10 years. That is to say that you should NOT rely on your home folder on the clusters for permanent storage. When you finish a computation, you should copy your results back to your own local computer(s) and make sure you have proper backups.

Additionally, CIBR operates a fairly large RAID array which has a complete replicated backup, making it dramatically safer. Any user may get a folder on this shared storage server, which can be accessed only from the head nodes on the clusters. Data stored on this RAID cannot be used for active processing, and access to the outside world is only through the head node on the clusters. It can be used to store files which you don't immediately need for processing, but expect to need again in the near future. It can also be used for short term backups of your processing results.

To get an allocation on the storage server, please contact the cluster system administrator.

2017 Update - We decided to mount the storage server directly on both Prism and Sphere, such that it is accessible from all nodes. However, access will be slower than the inbuilt RAID, and if the storage node goes down, it will impact your running jobs. We still recommend that most users use the inbuilt RAID on each cluster for most of their active processing.

CIBRClusters/Store (last edited 2018-03-11 04:27:23 by SteveLudtke)