The El Capitan systems mount the usual LC File Systems and users can familiarize themselves with the Using LC File Systems guide. Additional information is discussed in the Livermore Computing Resources and Environment tutorial, including details on
- quotas
- backups
- purging
- file transfer tools
Rabbit Storage
The El Capitan systems consist of a novel near node local storage technology called Rabbit. At its core, a Rabbit node consists of 16 SSDs and one AMD Epyc storage processor. Two Rabbit nodes sit in each El Capitan system rack and each is directly connected to 16 compute nodes within the same rack. The Rabbit nodes are also connected to the network to allow for cross-rack storage configurations. Combined with flux orchestration, there are a wide range of use cases. Future uses of the Rabbit storage system will allow for dynamic storage allocations, with cross-node storage available on a per-job basis.
Addition information about the Rabbits is linked below.
- https://www.hpcwire.com/2021/02/18/livermores-el-capitan-supercomputer-hpe-rabbit-storage-nodes/
- https://www.youtube.com/watch?v=xccViZtVye4
Static Storage Allocation
We have made a static, local storage allocation available to jobs on the CORAL 2 systems, including El Capitan, Tuolumne, and RZAdams. This XFS-type file system is mounted at /l/ssd on batch (not login) nodes and provides approximately 200GB of space available on a node-local basis (meaning files written from compute node are not directly visible to other compute nodes in the allocation). At the end of an allocation all data in /l/ssd is removed.
Dynamic Storage Use Cases
Friendly users are invited to utilize dynamic Rabbit allocation functionality. Documentation and flux allocation script examples can be found on the Flux with Rabbits for CORAL 2 page. Both XFS and Lustre storage allocation modes are available for user testing.
Early Users
Through the Rabbit storage nodes, we will eventually offer GFS2-type file system allocations along with the ability to run user-deployed containers. If you are keen on testing these features or working with Rabbits, please reach out to Elsa Gonsiorowski <gonsiorowski1@llnl.gov>.