media.ccc.de: MRMCD2026 - Storing PBs of data, at GigaBytes per second

https://media.ccc.de/v/2026-620-storing-pbs-of-data-at-gigabytes-per-second

The LHC generates lots of data, we will show case 1 system dCache that is used to store it all, and how we tune the boxes to be able to serve files as fast as possible.

The multiple experiments at the LHC generate a lot of data, and this all needs to be stored but also retrieved for processing. One of the main software storage suites used for this is [dCache](https://dcache.org/). A small deployment often already goes in the multiple PBs of storage. dCache only managed the data storage, (possible replication) and access via multiple protocols, (for examplenfs, webdav, xrootd, gridFTP), and behaves similar to to S3 in that sense that files are create and append, no random writes.

Tuning the underlying RAID configuration, kernel configuration and other settings within dCache to make sure we can reach line rate (or as close to line rate as possible) requires quite some testing and tuning.

We will over cover the basic architecture of dCache, the special protocols and challenges that exist in High Energy physics computing, and how to tune the kernel / filesystem to optimally handle this interesting load.

Evey

https://talks.mrmcd.net/2026/talk/7GRZ7E/

#mrmcd26 #InfrastructureandSoftware

https://creativecommons.org/licenses/by-sa/4.0/