Our future EOS installation will be around 3 PB, out of 9 FSTs. Since it is a new installation we’ll go with quarkdb (3 mgm nodes). We’ll be hosting CMS, Alice and Belle2 data, but we have no prior experience with running Grid services, including storage.
What I’ve seen in the documentation of quarkdb, SSDs are very much recommended.
I understand that the metadata store size depends more on the number of files, than their size, but are there some general recommendations on how much disk space to calculate for quarkdb store?
In our instances we see QDB disk usage of around 0.1 - 0.2GB per million files. The actual figure may vary depending on things such as how long or compressible the filenames are, but it seems quite consistent across all our instances.
Note there is a fixed size for the raft journal, and on default configuration it will reach around 20-40GB. The journal will eventually grow to be that size regardless of how many files there are in the namespace, so early on you might get a higher value than 0.1 - 0.2 GB per million files if you include the size of the raft-journal in your calculations.