[Bioclusters] Urgent advice on RAID design requested

Angulo, David dangulo at cti.depaul.edu
Thu Jan 18 09:55:59 EST 2007

you say your point stands.  I say it does not.  Please compare the actual MTBF figures.

-----Original Message-----
From: bioclusters-bounces+dangulo=cti.depaul.edu at bioinformatics.org on behalf of Tim Cutts
Sent: Thu 1/18/2007 1:56 AM
To: HPC in Bioinformatics
Subject: Re: [Bioclusters] Urgent advice on RAID design requested

On 17 Jan 2007, at 5:50 pm, Angulo, David wrote:

> I disagree with you.  Look at http://www.bestbuybusiness.com/bbfb/ 
> en/US/adirect/bestbuy? 
> cmd=catProductDetail&productID=BB10725830&operation=showDetails   
> This is RAID, so the mean time between failure will be extremely  
> large.  This is extremely reliable.

RAID makes things *more* reliable, yes.  "Extremely reliable" is a  
strong statement, and given our practical experience with RAID  
systems at various price points here, I would say somewhat optimistic.

That particular unit is SATA, so the MTBF of the individual spindles  
is quite high compared to higher end products.  When one of the  
spindles fails, you then have a period of vulnerability until the  
spindle is replaced, and the array rebuilt.  During that period, your  
probability of a second disk failure is much higher with SATA than it  
is with more expensive disks.  So your probability of data loss is  
still higher than it is with more expensive disks.   0.01 * 0.01  
might be a low probability, but 0.005 * 0.005 is a lot lower (yes,  
I've made those numbers up, but the point stands)

Secondly, you're forgetting that every time a spindle fails that's  
administrative work to replace it, so especially once your storage  
grows large, it's a significant workload to replace the disks.  If  
you're only buying one of these units, that's probably OK.

Thirdly, the performance of that little NAS server from multiple  
clients is probably not going to be that good, especially for the  
things that the original poster was mentioning using it for.

As I say, it's more a case of cost-benefit analysis.  How important  
is the safety of the data?  How important is performance under  
parallel load?  How much administrator time do you have to look after  
the kit?

I admit that little box does look rather nice though - I might buy  
one for home, where the cost-benefit is definitely in its favour.

Bioclusters maillist  -  Bioclusters at bioinformatics.org

More information about the Bioclusters mailing list