Silenus is a high performance distributed shared scratch storage infrastructure available from the Dahu and Bigfoot clusters.
This storage infrastructure has been set up to offer a high performance storage space with lowest possible latency and highest possible throughput, in particular for I/O intensive operations.
Silenus currently uses BeeGFS on a full SSD storage infrastructure with NVMe controllers.
Hoyt is the same technology of Silenus, but only available from the Kraken cluster. It is fully based on NVMe storage and plugged on the 200Gb/s infiniband network of Kraken, using RDMA transfers.
Silenus is only available on the Dahu and Bigfoot clusters as a local filesystem mounted on /silenus
.
Hoyt is only available on the Kraken clusters as a local filesystem mounted on /hoyt
.
The good practise is to use the directories created for each project under /silenus/PROJECTS/
or /hoyt/PROJECTS
which allow sharing data with all members of a project.
Moreover, in each project directory, subfolders are automatically created. A COMMON
folder, read and write accessible by all project members as well a a subfolder per project member, read only accessible to other project members.
This hierarchy offers project members a shared, secure and structured storage space.
Data storage on the Silenus/Hoyt infrastructure outside of the /[silenus|hoyt]/PROJECTS/
hierarchy is not encouraged.
There are also /[silenus|hoyt]/chunks_*
directories, each directory having two values following in its name.
These directories automatically distribute data in blocks of different sizes (the second value, 1MB or 512KB) over a fixed number of resources (the first value, 4, 8 or 12 storage resources).
If your jobs are able to make use of parallel access to data blocks of defined size, you can use these folders to automate the distribution of your data on Silenus hardware resources according to the parameters defined for each folder.
If you have any questions about the usefulness of these folders for your data, do not hesitate to contact us at sos-calcul-gricad@univ-grenoble-alpes.fr.
Please note that Silenus and Hoyt are scratch space. All data over 30 days old will be automatically deleted. You will erceive a reminder email listing data over 20 days old, allowing you to recover them before they are removed. However, we strongly encourage you to transfer your data to a more durable storage space immediately at the end of your job.
Moreover, there is a /[silenus|hoyt]/0-USAGE/users/
folder in which you will find two files per user. These files contain a daily recap giving, for each user with data on the infrastructure, the folders holding the greatest number of files and the most amount of data. These files are only accessible to the user they concern.