Ceph: 4 Nodes + LSI 630j Jbod (12 disks) for Bacula VM only

[rbg]

hi,

I got new 4 IBM V3700 RBODs and they are not compatible with our SAS Switch (LSI 6160) :-( So i have to move all VMs from our "old" environment to the new one. After that, I have 4 x LSI 630J JBODs with 12 x 2TB SAS (Seagate Constallation) free. My idea was to connect one JBODs to our 4 ProxMox nodes (3.2) (one JBOD/node) and create a CEPH Cluster over all 4 nodes.

The nodes are IBM 3550M3 (64GB DDR3) with two SAS 300GB SAS disks, configured as RAID1 and ProxMox installed on it. They has also two 1 Gib/s Interfaces, for internal communication (ISCSI and ProxMox related things) (no bonding, just multipath for ISCSI) and cannot upgraded at this time (no more free slots). I would buy 4 LSI 9200-8e SAS hostbusadapters and use them for the JBODs. I want them only for my Bacula/Bareos VM (96 TB RAW capacity minus CEPH and filesystem) as backup to disk target.

Could be the 2 x 1Gib a real problem? As it is "only" backup, and at the moment, the backups are written over ISCSI. But, 96TB ...

any suggestions?

 

[udo]

Hi,
you should test your config.

For an backup-server you need fast IO... and IO can be an bottleneck with kvm...

I have two bacula-server inside an CT to avoid IO-problems - all necessary filesytems are done with bind-mounts (postgres-database, spool-area and backup-storage).
But I use only small space for backups (10TB) because the main backups are stored in the lto-jukebox.

Udo

 

[rbg]

hi Udo,

thanks for the reply and sorry, for the very late answer my goal is the keep most parts identical, so the backup VM too. Everything is KVM which makes the documentation easier , but, however the IO is at this time not a real problem, because everything can be "backuped" in a few hours.
Also the the bottleneck is the 1Gib/s connection to the outside.
Before I move to ceph, I want to test it with the backup KVM VM and see, if and how it is working

cu denny