Cta objectstore on shared NFS and objectstore CEPH size

Hi all!

While setting up eoscta prototype I faced stage about CTA objectstore.

Are there any known limitations to the objectstore on a file system which is shared over NFS? Has ever somebody tested such setup on small instances. We have two tape drives with one 80 LTO8 tape slots library. And whole eoscta uses three physical servers. My simple tests on the real hardware work (eoscta archive, retrive files) But can not find any information about this feature (file objectstore in preprod) in available docs and presentations.

And I have another question about objectstore size. Are there any estimations how big it could be if CEPH is used?
Thx in advance!

Hi Victor,

At CERN, we have only used the file-based objectstore for development/testing. That’s not to say that it won’t work, but we never tested it.

I seem to remember that one of the other sites may have done this, if you don’t get a reply here, check in the site presentations from the EOS Workshop 2022.

The size of the objectstore is pretty small. It’s a function of the number of items you have queued at one time, but one queued request is only a few hundred bytes. Even with millions of items queued it shouldn’t exceed a few GB.


hi Victor,

Just to chime in, we are using NFS as an objectstore since issue #1069 was resolved and so far we haven’t observed any setbacks. Whether on a small or large instance, it pretty much works as expected.


Hi, Michael, Mwai!

Many thx! So we will stay with NFS. To simplify software stack.