[ILUG] How useful are NAS devices aimed at home users?

Darragh Bailey felix at compsoc.nuigalway.ie
Thu Nov 20 17:27:43 GMT 2008


On Thu, Nov 20, 2008 at 04:53:03PM +0000, Michael Watterson wrote:
> You'd need to be using a lot more than 4 -5 disks to justify RAID6. It's  
> for VERY big arrays where it's possible that a 2nd drive might fail  
> before rebuild.

If I end up with 4 the RAID 1 is fine, if 5, then RAID 6 makes more
sense as I only lose 2 disks to raid. With 1TB sata disks the odds of 
hitting an unrecoverable read error during a raid 5 rebuilt is rather 
high, and will continue to increase as disk sizes increase without URE 
rates reducing for SATA disks.

> SATA's goodness or not for RAID5 or any RAID other  than Mirror, depends  
> on the interface Hardware and Drivers.
>
> SCSI works well for RAID5 as it can do concurrent operations of every  
> drive. If the SATA or IDE based RAID controller has a separate port for  
> each drive than can be true for it too.
>
> There are plenty of these NAS that can be customised. Note the CPU  
> performance seriously limits applications.
>
> Very few do RAID5
>
> http://www.cnet.com/topic-reviews/nas/server.html

Odd, a quick look at komplett suggests that any of the 4-5 bay (non
rack) devices <€800 all do raid 5 and some have raid6

> *Western Digital  ShareSpace Home NAS Array*
> This is the only one I know that does RAID5 out of the box. Most only  
> support Mirror (RAID 1)  or Stripeset (RAID0) with no redundancy using a  
> pair of drives.
>
> Four drives is a reasonable RAID5 system, though my attic UltraWide &  
> Fast 10,000rpm SCSI box has 5 drives for data and 3 for OS on two  
> separate controller channels for speed.
> http://www.computerworld.com/action/article.do?command=viewArticleBasic&taxonomyId=154&articleId=9116079&intsrc=hm_topic

4-5 disk RAID 5 with SCSI disks is fine, but not with SATA, the story is
not so good.

SATA drives don't have as low error rates as SCSI disks, and with the
size of the SATA disks increasing, the chances of an error occuring
during a rebuilt is getting uncomfortably higher.

http://blogs.netapp.com/dave/2006/03/expect_double_d.html

This one spells it out much clearer
http://blogs.zdnet.com/storage/?p=162

With TB disks, I might only have 4TB of data, but that means I still
have about 33% chance of hitting an URE during a rebuild. i.e (numbers
taken from the article) SATA disks have URE numbers of 10^14, which is
one error every ~12TB of data. During a rebuild, the 4TB of data will
have to be read to populate the new disk, so there's ~1/3 probability
that an URE will hit.

Sorry but that's not a nice number. So basically, once you get to 2-3TB
of data protected by a RAID array, you'll want to be able to survive a 2
disk failure, not just 1, unless of course you like seeing corrupted
array's.

Also I've some experience at work that suggests that the guy is right.
With arrays of 12 Disk's of 300GB in size, we've seen URE's occur on
sata disks when rebuilding arrays. This is why we won't use RAID 5 with
sata disk's any more, the URE numbers are just too high relative to the
data size of the array's that you are dealing with.

> Most Home NAS will have disappointing performance due to x86 Geode, MIPS  
> SoC or similar CPUs.

Not too worried about performance if I manage to get everything setup
correctly. Caching some of the data locally should avoid too many
problems.

> We have experimented with sleeping our server and using WoL. However we  
> use it as a Mail Server 24x7 picking up & forwarding every few minutes.

I doubt I'll be running it quite as often :)

> -- 
> Mike

-- 
Darragh

"Nothing is foolproof to a sufficiently talented fool."



More information about the ILUG mailing list