[Gluster-users] Tiered Volumes and Backups

Dan Lambright dlambrig at redhat.com
Mon Aug 8 14:01:53 UTC 2016

----- Original Message -----
> From: "Mohammed Rafi K C" <rkavunga at redhat.com>
> To: "Lindsay Mathieson" <lindsay.mathieson at gmail.com>, "gluster-users" <Gluster-users at gluster.org>
> Sent: Monday, August 8, 2016 1:03:19 AM
> Subject: Re: [Gluster-users] Tiered Volumes and Backups
> On 08/07/2016 08:41 AM, Lindsay Mathieson wrote:
> > On 7/08/2016 12:34 PM, Lindsay Mathieson wrote:
> >> One of the things done every week here is a full image backup of all
> >> the VM's (take about 20 hours).
> Do you use gluster snapshot feature here ? If so, it won't result in
> promotion.
> >> I was wondering what the implications are for the tier as every shard
> >> will be read sequentially. Will I be seeing mass promotions and
> >> demotions into the Tier? this would not be desirable.
> >
> > I'm guessing cluster.tier-promote-frequency and
> > cluster.read-freq-threshold would be the relevant settings to control
> > this?
> Yes. You can control the behavior using the options. Also we are
> planning to give a pause or stop option for migration daemon so that you
> can temporarily suspend migration.

By default, files are marked for promotion on the first I/O. So if you did a backup and touched every file, they would all be marked for promotion. You ought to be able to change that, so only files touched some number of times (2X?) are promoted. See the "cluster.write-freq-threshold" setting below.

What size files are you using? The way things work (right now), the smaller the file, the less benefit you will see... the cost of metadata round trips (LOOKUPS) start to exceed faster data transfers from SSDs. Many efforts are going into trying to fix that... this same problem may be holding back benefits from RDMA and other areas as well.

Option: cluster.write-freq-threshold
Default Value: 0
Description: Defines the number of writes, in a promotion/demotion cycle, that would mark a file HOT for promotion. Any file that has write hits less than this value will be considered as COLD and will be demoted.

> Regards
> Rafi KC
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://www.gluster.org/mailman/listinfo/gluster-users

More information about the Gluster-users mailing list