[Gllug] Cheapish NAS.

Nix nix at esperi.org.uk
Wed Feb 17 00:38:50 UTC 2010


On 16 Feb 2010, JLMS told this:

> On Tue, Feb 16, 2010 at 4:01 AM, general_email at technicalbloke.com
> <general_email at technicalbloke.com> wrote:
>> Option 2. A small PC based server with a 3 disk software RAID 5,
>> possibly running FreeNAS or OpenFiler. Given that I don't really want
>> the hassle I'd want to charge at least £100 for my time building &
>> testing the thing. That would leave ~£350 for all the bits. Does that
>> seem realistic for a box with reasonable performance (say avg 40MB/s
>> throughput)?
>
> I will not even try to address the throughput, I have never had to
> care much about it, normally what has been provided to me has been
> good enough (performance was not an issue, security and availability
> were), in any case, £350 sounds awfully little, that would be aiming
> for one of the most basic machines you can find, so most likely the
> quality of the components will be on the lacking side of things. If on
> top of that you put RAID5 (which is notoriously inefficient for
> writes) then you are asking for trouble.

It may be notoriously inefficient, but I've had md/RAID-5 systems
clocking 210Mb/s (with four PCIe SATA drives). They're only really
inefficient if you're doing a lot of random writes with blocksizes not
aligned to the RAID array's stripes: so as long as your fs is aligned to
the array stripes (automatic with recent e2fsprogs and LVM), it should
just work. Plus, most writes aren't random unless you're running a
database, and most I/O is not writes.

More of a concern with RAID-5 and RAID-6 is that if you lose power in
the middle of a write you can expect to see disk corruption, so if power
failures are more common than drive failures, software RAID-5/6 is a bad
idea. (Battery-backed arrays avoid this problem, because all writes
initiated to any stripe are guaranteed to complete to all stripes:
they'll never fail without marking the array as failed, rewriting, or
otherwise attempting some form of recovery.)

> Disks are so cheap that for this solution I would buy  2 as big as
> affordable and do a mirror, avoiding the performance penalty of RAID5.

... and chucking a lot of disk capacity. (Mind you, modern disks are so
huge this may not be at all important.)

I have one (hardware) RAID-5 array at home and one RAID-1. I don't keep
*anything* non-RAIDed anymore: disks are so cheap that the worry isn't
worth it. (Well, one flashcard is non-RAIDed, but the master image is on
the RAID-5 array, synched nightly.)

>> I've never built a raid based system before, how much more
>> to recovering is there than just slinging all the drives in a fresh box
>> and hitting the power switch if, say, the mobo suddenly died? Pls bear
>> in mind I'm no linux ninja, almost all my work is supporting Windows &
>> Mac at the domestic level and I just switched my workshop machine to
>> Ubuntu a mere 18 months ago!
>
> Uhm. You don't do that (I don't even know if it is possible). You redo
> your RAID5 volume and recover from archival or backups. Much faster in
> general terms.

Why? I did exactly that when I lost my firewall to the God of Static.
As long as the kernel for the old machine works on the new one, you're
fine: so keep a generic kernel around in your boot menu for such
situations (not necessarily the default).

It's a hell of a lot easier to get the same disk image you had trouble
with back than it is to rebuild the thing from a backup, even if you
do nightly backups.
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug




More information about the GLLUG mailing list