Re: Enhanced Synology CS/RS support

[David Brownlee <abs%netbsd.org@localhost>]

On 19 April 2012 13:17, Frank Wille <frank%phoenix.owl.de@localhost> wrote:
> On Thu, 19 Apr 2012 12:26:27 +0100
> David Brownlee <abs%netbsd.org@localhost> wrote:
>
>> [...] and the performance difference
>> between a 3 or 5 drive RAID5 and a 4 drive RAID5, at least on
>> RAIDframe can be massive.
>
> Interesting. I didn't know that.
> Then the difference would be even bigger on a NAS with a weak MPC8245 CPU.

I'm not so sure its CPU as IO bound. Generally the best performance
comes from a filesystem block size
which matches the strip size ("generally"), but if the stripe size is
a non power of two then either the filesystem block size needs to be a
power of two which fits into it an integer number of times, or we hit
read-write-modify hell .

>> I'm currently using a couple of HP microservers with 5*2TB disks under
>> NetBSD/amd64, but I'm always on the lookout for something neater :)
>>
>> Actually - 5 or more would work for me, though for an 8 I'd probably
>> end up using a 5 + 3...
>
> QNAP makes nice NAS. I would recommend something like a QNAP TS-509Pro or
> a TS-559Pro. Both are x86-based (Celeron and Atom), but I don't know how
> well NetBSD would work on it.

I wonder how they compare powerwise to an HP microserver - which is
around 150 GBP after cashback...

I have two of the earlier model of http://www.ebuyer.com/281915 and
apart from needing an adaptor to put
the 5th HD into the optical media bay its pretty much ideal - even
down to a pci-e slot for a second network card...

Still, I'm getting away from port-sandpoint relevance :) I'd be quite
curious to see how non power-of-two+1 RAID5 configurations stack up on
the NAS boxes...