PostgreSQL server embedded in NAS firmware?

Scratch that. An immediate show stopping pitfall occurs to me: the
necessity to match CPU/OS Architecture between primary server and replicate
target. Doubtful that there are any consumer NAS products out there
running linux on 64bit/intel

On Fri, Sep 7, 2012 at 8:23 AM, Andrew Barnham <andrew.barnham@gmail.com>wrote:

Andrew Barnham <andrew.barnham@gmail.com> writes:

Scratch that. An immediate show stopping pitfall occurs to me: the
necessity to match CPU/OS Architecture between primary server and replicate
target. Doubtful that there are any consumer NAS products out there
running linux on 64bit/intel

Maybe not, but there are with 32-bit Intel ... if you really want to do
this, there's nothing to stop you from running a 32-bit build on your
primary machine and then replicating to the NAS. This would limit what
you could crank shared_buffers up to, but otherwise should work fine.

regards, tom lane

That shouldn't really matter. Either the db is just on the NAS in
which case as long as pg compiles on it then the client on the main
unit shouldn't matter, or the data is just stored there and the db is
on the main unit, client and all and again it wouldn't matter.

But the client and server do NOT have to be the same architecture to
work for sure.

On Thu, Sep 6, 2012 at 4:40 PM, Andrew Barnham <andrew.barnham@gmail.com> wrote:

Scratch that. An immediate show stopping pitfall occurs to me: the necessity
to match CPU/OS Architecture between primary server and replicate target.
Doubtful that there are any consumer NAS products out there running linux on
64bit/intel

On Fri, Sep 7, 2012 at 8:23 AM, Andrew Barnham <andrew.barnham@gmail.com>
wrote:

Hi

I currently run a modest streaming replication target on a cheap, single
disk ASUS media center; replicating a 100GB PG database.

I want to add RAID via a consumer grade NAS device.

As far as I can tell consumer grade NAS devices these days appear to be
fairly rich & flexible embedded lniux/freebsd systems.

Has anyone had any experience with running postgresql on the NAS device
itself? Which products? Any traps or pitfalls or integrity concerns about
such an arrangement?

Andrew

--
To understand recursion, one must first understand recursion.

On 09/06/2012 04:19 PM, Scott Marlowe wrote:

That shouldn't really matter. Either the db is just on the NAS in
which case as long as pg compiles on it then the client on the main
unit shouldn't matter, or the data is just stored there and the db is
on the main unit, client and all and again it wouldn't matter.

But the client and server do NOT have to be the same architecture to
work for sure.

If I understood the OP, it is not client <--> server, it is:
main server <--> replication server

In that case architecture would matter.

On Thu, Sep 6, 2012 at 4:40 PM, Andrew Barnham <andrew.barnham@gmail.com> wrote:

Scratch that. An immediate show stopping pitfall occurs to me: the necessity
to match CPU/OS Architecture between primary server and replicate target.
Doubtful that there are any consumer NAS products out there running linux on
64bit/intel

On Fri, Sep 7, 2012 at 8:23 AM, Andrew Barnham <andrew.barnham@gmail.com>
wrote:

Hi

I currently run a modest streaming replication target on a cheap, single
disk ASUS media center; replicating a 100GB PG database.

I want to add RAID via a consumer grade NAS device.

As far as I can tell consumer grade NAS devices these days appear to be
fairly rich & flexible embedded lniux/freebsd systems.

Has anyone had any experience with running postgresql on the NAS device
itself? Which products? Any traps or pitfalls or integrity concerns about
such an arrangement?

Andrew

--
Adrian Klaver
adrian.klaver@gmail.com

On Fri, Sep 7, 2012 at 4:10 AM, Andrew Barnham <andrew.barnham@gmail.com>wrote:

Scratch that. An immediate show stopping pitfall occurs to me: the
necessity to match CPU/OS Architecture between primary server and replicate
target. Doubtful that there are any consumer NAS products out there
running linux on 64bit/intel

FreeNAS is based on FreeBSD 8.2 and is available in 64 bit arch.

Amitabh

On Thu, Sep 6, 2012 at 6:06 PM, Adrian Klaver <adrian.klaver@gmail.com> wrote:

On 09/06/2012 04:19 PM, Scott Marlowe wrote:

That shouldn't really matter. Either the db is just on the NAS in
which case as long as pg compiles on it then the client on the main
unit shouldn't matter, or the data is just stored there and the db is
on the main unit, client and all and again it wouldn't matter.

But the client and server do NOT have to be the same architecture to
work for sure.

If I understood the OP, it is not client <--> server, it is:
main server <--> replication server

In that case architecture would matter.

Ahh I thought he'd be moving both ends of the replication onto embedded nas.

On 6 September 2012 23:40, Andrew Barnham <andrew.barnham@gmail.com> wrote:

Scratch that. An immediate show stopping pitfall occurs to me: the necessity
to match CPU/OS Architecture between primary server and replicate target.
Doubtful that there are any consumer NAS products out there running linux on
64bit/intel

Hi

I have a super cheapskate rig along those lines at home, doing
replication among other things: I used an HP Microserver (not
marketed as a 'consumer NAS' exactly, but the same general idea:
a low cost black cube with drives bays, SATA ports, a small
amount of ECC RAM and a low power dual core amd64 CPU). I run
Debian GNU/Linux and have a bunch of PostgreSQL databases,
backups and virtual machines on it. My goals were: cheap to buy,
cheap to run, reasonably reliable, quiet, small, inoffensive to
the eye. I filled it up with 'green' 5400RPM drives that I had
spare from another project, configured software RAID arrays with
XFS on top, and put it on a shelf to run headless. A friend has the
same box but runs FreeNAS on it so he can use ZFS and swears by
it (he also added a 4 x 2.5" adaptor to be able to reach the
maximum of 8 drives, which I think requires adding a controller
card, whereas I used the 5.25" bay for a 5th 3.5" drive). The
machines were going for around 150 GBP when I bought, and I added
some RAM. Last time I measured it it was drawing around 50W (a
bit more when busy, a bet less when idle), which works out to
under 50 quid a year to run at London retail electricity prices,
comparable to a light bulb. This is surely about the slowest
database hardware money can buy, but handles my hobbiest
databases (~1TB for the largest) and a bunch of streaming
replicas and backups from remote servers just fine. I haven't
checked, but I would expect it to be the slowest build farm
member...

Thomas Munro