Hi Cephers,
Our university could deploy ceph. The goal is to store datas for research laboratories (non-HPC).
To do this, we plan to use Ceph with RBD (mount block device) from a NFS (or CIFS) server (ceph client) to workstations in laboratories.
According to our tests, the OS (ubuntu or centos...) that map the RBD block implements file system write cache (vm.dirty_ratio, etc ...). In that case, the NFS server will always perform writes to workstations whereas it has not finished writing datas to Ceph cluster - and regardless of whether the RBD cache is enabled or not in the config [client] section.
My questions:
- Does the activation of RBD cache is useful only when it combines Virtuals Machnies (where QEMU can access an image as a virtual block device directly via librbd) ?
- Is it common to use Ceph, with RBD to share network file systems ?
- And if so, what are the recommendations concerning the OS cache ?
Thanks a lot.
Stephane.
--
Université de LorraineStéphane DUGRAVOT - Direction du numérique - Infrastructure
Jabber : stephane.dugravot@xxxxxxxxxxxxxxxx
Tél.: +33 3 83 68 20 98
_______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
