Gluster users - Nov 2015 - Getting the best performance v lowest energy use and small form factor

Hi,

Thanks but, your solution doesnt protect for a single PC hardware failure
like a PSU blowing ie giving me real time replication to the 2nd site so I
can be back up in minutes.

On 4 November 2015 at 10:41, Lindsay Mathieson <lindsay.mathieson at
gmail.com>
wrote:
>
> On 4 November 2015 at 06:37, Thing <thing.thing at gmail.com> wrote:
>
>>
>> Looking at running a 2 node gluster setup to feed a small
Virtualisation
>> setup.  I need it to be low energy use, low purchase cost and small
form
>> factor so I am looking at 2 mini-itx motherboards.
>>
>> Does anyone know what sort of throughput I can expect? ie I am looking
>> for as good as a single drive in the VMare/kvm box but simply having
>> redundancy for data protection and availability so nothing huge is
required.
>
>
> Well read throughput is going to depend entirely on your drives and
> network, write through put almost certainly bound by your network. It
> probably won't be as good as a single drive by itself - realtime
> replication by its nature is going to be slower..
>
> For your sort of setup I'd be looking at a ZFS setup, RAID1 or 10 if
you
> can afford it with a SSD read/write cache (easy with ZFS), it will have
> pretty good performance and lots of extra goodies such as snapshots,
> compression (very good), checksuming/bitrot detection. That in and of
> itself with have data redundancy on the one PC.
>
> If you absolutely need redundancy across two PC's then you could setup
a
> 2nd ZFS server and use ZFS's builtin differential snapshots to
replicate to
> the 2nd server on a regular interval.
>
>
> --
> Lindsay
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://www.gluster.org/pipermail/gluster-users/attachments/20151104/a5815201/attachment.html>

On 4 November 2015 at 08:39, Thing <thing.thing at gmail.com> wrote:
> Thanks but, your solution doesnt protect for a single PC hardware failure
> like a PSU blowing ie giving me real time replication to the 2nd site so I
> can be back up in minutes.
>
ZFS can be configured to replicate every few minutes - whether that is
sufficient is dependant on your uptime and data loss requirements.

If you *must* have realtime redundancy then yes something like gluster or
ceph is your only option. Gluster is easier to setup and maintain then
ceph. Both of them are a lot more reliable if you have three nodes, two
nodes is asking for trouble - split brain etc.

If you want some throughput estimates then we need more spec's:

- RAM
- CPU
- Hard Disks
- Network
- Overall Config
  * Caching
  * Bonding
  * etc

- With a std 1GB ethernet, your writes will max out at around 110 MB/s
- Same for Reads, unless your VM Host is also your gluster node, in which
case your reads will be a bit slower than your underlying file system
access times


-- 
Lindsay
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://www.gluster.org/pipermail/gluster-users/attachments/20151104/feb2c89b/attachment.html>