I read the Wiki at https://pve.proxmox.com/wiki/Ceph_Server and their config doesn't show a RAID controller. If I where to use a Dell r710 or a r420 would I simply put the RAID controller in passthrough mode or not have one at all?
You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an alternative browser.
You should upgrade or use an alternative browser.
You should have some type of redundancy for the system on which you will deploy Ceph, you can use RAID controller to do so. The actual Ceph storage will be on jbod and not RAID as the Ceph will create redundancy for the drives without needing a dedicated controller.
So.. My storage controller will be in JBOD mode which means there will be no RAID enabled. That is basically pass through. So, storage redundancy of my Ceph objects will be created by the ceph cluster system. I thought about using a SSD for my system drive and another SSD for my journal. That should be okay? Or should I have those on a mirror or raid 10?
Last edited:
What I am saying is that you deploy Ceph on some OS, you can use Ubuntu or ProxmoxVE to do this. This OS install should have some redundancy like RAID or ZFS. Once you deploy Ceph on it, Ceph will store the data on other hard drives that are not a part of a RAID array as Ceph creates its own redundancy I think by default it is 2 replicas.
In the guide located at https://pve.proxmox.com/wiki/Ceph_Server in the recommended hardware section they have a SSD for the OS and another SSD for the journal. The OSD's for ceph reside on the ceph storage cluster. Do you think I should steer away from this config and use a mirror for the OS and Journal? That way I can get a 6 x bay server. 4x for ceph storage cluster and 2 x for mirror. I like their approach for it uses fast SSD drives. But... They are not redundant
Last edited:
I'm using dell r630 for my ceph cluster, with H330 sas controller.
This controller can mix raid and passthrough.
I'm using raid1 for os + ceph monitor, and passthrough for ceph osds.
This controller can mix raid and passthrough.
I'm using raid1 for os + ceph monitor, and passthrough for ceph osds.
I'm using dell r630 for my ceph cluster, with H330 sas controller.
This controller can mix raid and passthrough.
I'm using raid1 for os + ceph monitor, and passthrough for ceph osds.
I can't decide which direction to go.. I really like the idea of a SSD booting the OS and another SSD for the journal due to write/read being so fast.. But. I'm concerned that it's not redundant. In the Ceph Server guide they recommend the SSD route.. I guess I need to flip a coin and decide?!
Why can't you make the SSDs redundant ? If you have controller that allows you to mix RAID and pass-through you should be able to do this. Definitely do not put the Journal on non redundant drives.
Hi,I can't decide which direction to go.. I really like the idea of a SSD booting the OS and another SSD for the journal due to write/read being so fast.. But. I'm concerned that it's not redundant. In the Ceph Server guide they recommend the SSD route.. I guess I need to flip a coin and decide?!
on my ceph cluster (not pve - pure debian) I use an "normal" SSD for the OS - deployed with puppet. And two SSDs for journals (also not raid-1).
The journal SSDs are good Intel DC S3700 and all SSDs are monitored by icinga.
If one host die (OS-SSD or anything else), ceph has enough redundance (replica=3) - so I'm not in trouble... due puppet the host is installed fast again and the journal-SSD are very very robust.
Udo