All of lore.kernel.org
 help / color / mirror / Atom feed
* Add OSD and hard drives
@ 2011-06-29 19:23 Hemingway, Graham Stuart
  2011-06-30  1:25 ` Josh Durgin
  0 siblings, 1 reply; 3+ messages in thread
From: Hemingway, Graham Stuart @ 2011-06-29 19:23 UTC (permalink / raw)
  To: ceph-devel@vger.kernel.org

Hello,

I have a simple ceph installation working.  My cluster has 7 hardware nodes.  5 OSDs (1 2TB drive each) and two MDSs.  I have one monitor which is on the same machine as one of the MDSs.  Running on Ubuntu 10.10 and v0.30 I have things working.  Yeah.
So, my questions are:

1) Would you recommend adding a second monitor on the same machine as the other MDS?  Therefore two nodes each with MON and MDS.  Does ceph still want monitors in odd numbers?
2) What is the procedure for adding an additional hard drive to each OSD?  I have a second 2TB drive physically on each OSD node that I would like to bring up and I did not find ready documentation on just adding drives.

Thanks,
   Graham

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: Add OSD and hard drives
  2011-06-29 19:23 Add OSD and hard drives Hemingway, Graham Stuart
@ 2011-06-30  1:25 ` Josh Durgin
  2011-06-30 12:45   ` Wido den Hollander
  0 siblings, 1 reply; 3+ messages in thread
From: Josh Durgin @ 2011-06-30  1:25 UTC (permalink / raw)
  To: Hemingway, Graham Stuart; +Cc: ceph-devel@vger.kernel.org

Hi Graham,

On 06/29/2011 12:23 PM, Hemingway, Graham Stuart wrote:
> Hello,
>
> I have a simple ceph installation working.  My cluster has 7 hardware nodes.  5 OSDs (1 2TB drive each) and two MDSs.  I have one monitor which is on the same machine as one of the MDSs.  Running on Ubuntu 10.10 and v0.30 I have things working.  Yeah.
> So, my questions are:
>
> 1) Would you recommend adding a second monitor on the same machine as the other MDS?  Therefore two nodes each with MON and MDS.  Does ceph still want monitors in odd numbers?

Generally the monitors don't need much cpu or memory, and it's a good 
idea to run 3, so your cluster continues working if one fails. An even 
number doesn't improve reliability since a majority of monitors need to 
be active to make progress.

> 2) What is the procedure for adding an additional hard drive to each OSD?  I have a second 2TB drive physically on each OSD node that I would like to bring up and I did not find ready documentation on just adding drives.

The simplest way to add a drive is to start a new OSD for the new drive. 
Instructions for this are on the wiki:

http://ceph.newdream.net/wiki/OSD_cluster_expansion/contraction

Since your new drive is the same size as the others, you wouldn't need 
to adjust your crush map to get an even distribution of data.

The downside of running one OSD per disk is increased memory and cpu 
usage, but this shouldn't be a problem with a small cluster and few PGs.

Josh

^ permalink raw reply	[flat|nested] 3+ messages in thread

* Re: Add OSD and hard drives
  2011-06-30  1:25 ` Josh Durgin
@ 2011-06-30 12:45   ` Wido den Hollander
  0 siblings, 0 replies; 3+ messages in thread
From: Wido den Hollander @ 2011-06-30 12:45 UTC (permalink / raw)
  To: Josh Durgin; +Cc: Hemingway, Graham Stuart, ceph-devel@vger.kernel.org

Hi Graham and Josh,

On Wed, 2011-06-29 at 18:25 -0700, Josh Durgin wrote:

> > 2) What is the procedure for adding an additional hard drive to each OSD?  I have a second 2TB drive physically on each OSD node that I would like to bring up and I did not find ready documentation on just adding drives.
> 
> The simplest way to add a drive is to start a new OSD for the new drive. 
> Instructions for this are on the wiki:
> 
> http://ceph.newdream.net/wiki/OSD_cluster_expansion/contraction
> 
> Since your new drive is the same size as the others, you wouldn't need 
> to adjust your crush map to get an even distribution of data.

You still need to add it to your crush map, but you don't need to fiddle
with the weight of the OSDs ;-)

Just to make it clear for Graham, you do need to add the new OSDs to
your crush map as new devices, but the weight isn't needed since they're
all of equal size.

> 
> The downside of running one OSD per disk is increased memory and cpu 
> usage, but this shouldn't be a problem with a small cluster and few PGs.

Yes, I can confirm that. I currently have 40 OSD's and +10k PG's, this
is hitting my machines pretty hard. The machines are Atoms with 4GB of
RAM each and 4 OSD's (2TB) per machine.

Wido


^ permalink raw reply	[flat|nested] 3+ messages in thread

end of thread, other threads:[~2011-06-30 12:45 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2011-06-29 19:23 Add OSD and hard drives Hemingway, Graham Stuart
2011-06-30  1:25 ` Josh Durgin
2011-06-30 12:45   ` Wido den Hollander

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.