[zfs-discuss] SSD, Ceph Journal, ZVOL and trim :)

Phil Harman phil.harman at gmail.com
Wed Feb 24 20:44:56 EST 2016


Lindsay,

I’ve not tuned Ceph, but I wonder why you would choose a two disk RAIDZ1 over a simple mirrored vdev?

I’d look for a pair of 100GB-200GB SSDs with power loss protection and as many IOPS as you need / can afford.

   e.g. something like the Seagate 600 Pro works quite well if you disable cache flushing (it doesn’t need it)

Then you’ll want to tune your ZVOL blocksize to suite your workload. You might also consider primarycache=metadata.

Does Ceph support journals in files? If so, it might be a little simpler to manage.

I wouldn’t worry about provisioning and TRIM, but it’s worth getting something with a 5-year warranty :)

Phil


> On 24 Feb 2016, at 07:24, Lindsay Mathieson via zfs-discuss <zfs-discuss at list.zfsonlinux.org> wrote:
> 
> I know this has been asked before but there doesn't seem to be a definitive answer ...
> 
> Looking at making the management of my ceph journals a little easier and zfs zvols seem a good match. However ceph journals are *very* busy, 50 GB/Day is quite normal for a small ceph cluster like mine.
> 
> So what happens with trimming? is it needed? if so, how do you trim a ZVOL?
> 
> 
> Also - data integrity of a ceph journal is essential as well - does anyone see a problem with the following?
> 
> - 2 * 120GB SSD's in  RAID1
> - 3 * 20GB zvols for 3  Ceph Journals?
> - 60GB left free for over provisioning
> 
> Should the zvols be sparsely provisioned?
> 
> 
> thanks,
> 
> -- 
> Lindsay Mathieson
> 
> _______________________________________________
> zfs-discuss mailing list
> zfs-discuss at list.zfsonlinux.org
> http://list.zfsonlinux.org/cgi-bin/mailman/listinfo/zfs-discuss



More information about the zfs-discuss mailing list