[one-users] ONE best practices and cluster design

[Ruben S. Montero]

Hi
I think there are three main reasons to use clusters:

1.- First is to adapt your virtual resources to your hardware setup. For
example networks  or datastores that are connected to some hosts but not
others.

2.- Second, it may help you to better plan capacity allocation, for
example, by  grouping low performance resources in a cluster and assigning
there low priority VMs; or use ACLs to give access to a given cluster to a
given set of users.

3.- Finally, performance. As you said probably having different system
datastores is a good idea to balance IO from the hosts. There are two
different metrics to consider while planning your storage:
  a) Provisioning time. The time a cloning an image depends on both the
technology (shared FS vs LVM for example) but also on the disk format (e.g.
qcow2 vs raw). I'd make a couple of tests with multiple simultaneous clone
operations, this will stress your systems. Note that for this also the
Image datastore is important.
  b) VM IO performance. The actual disk IO performance when multiple VMs
are accessing their disk images. This may be only relevant if your are
interested in such workloads.

So a system DS using your NetApp SAN can be assigned to a cluster of hosts,
there is no need to create more than one cluster if they are using the same
SAN.

Then using the SSDs as a System DS could be also a good idea. You can share
the Image DS from the SAN and clone images locally in the SSD'd (e.g. using
TM SSH for this system DS). You'll have a great VM IO performance, but
consider the cloning time. Also you can put a distributed FS like Gluster
FS on top of the SSD's.....

Sorry for the long and vague mail, but there is no silver bullet for
this....Hope it helps,

Cheers

Ruben


On Sat, Sep 14, 2013 at 12:02 AM, Gabriel Cain <gabriel at moz.com> wrote:

> Hi folks.
>
> So I'm new the the mailing list, and fairly new to Open Nebula.  I've been
> working on understanding how to set it all up over the last month or so.
> I've got it where I can do all the things, and that's pretty cool.
>
> Where I'm at now is that I'm trying to figure out how to turn a system on
> a few nodes into a proper production cluster.
>
> Some background on the goals:
> * We're planning on moving out of EC2 for a lot of our dedicated compute.
> * I have a lot of powerful machines to work with -- 32 Dell blades (128G
> ram / 64 CPUs), 30 Dell C-220 servers (128G ram, 64 CPUs, 8 disk bays)
> * I have a few TB of Netapp backed NAS
>
> I'm not finding a lot of information on how to design such a cluster to
> properly perform.  I'm worried about disk IO on the NFS volumes.
>
> I'm leaning towards putting the blades on the NFS volume for their SYSTEM
> datastore, and doing SSDs for the C220s for their SYSTEM datastore.  I'm
> not sure I'm doing it right when it comes to those.
>
> Anyway, I'd appreciate some guidance on how to take an experimental
> cluster to being a production cluster such that I have high reliability and
> availability here.
>
> -Gabriel
>
>
> _______________________________________________
> Users mailing list
> Users at lists.opennebula.org
> http://lists.opennebula.org/listinfo.cgi/users-opennebula.org
>
>


-- 
-- 
Join us at OpenNebulaConf2013 in Berlin, 24-26 September, 2013
-- 
Ruben S. Montero, PhD
Project co-Lead and Chief Architect
OpenNebula - The Open Source Solution for Data Center Virtualization
www.OpenNebula.org | rsmontero at opennebula.org | @OpenNebula
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opennebula.org/pipermail/users-opennebula.org/attachments/20130917/0a23f494/attachment-0002.htm>