[Bioclusters] Urgent advice on RAID design requested

Malay mbasu at mail.nih.gov
Thu Jan 18 18:06:44 EST 2007


Joe Landman wrote:
> Angulo, David wrote:
>> you say your point stands.  I say it does not.  Please compare the 
>> actual MTBF figures.
> 
> Hi David:
> 
>   The MTBF of the system is related to the MTBFs of all of the 
> components.  If the MTBFs of the disks are so large that the power 
> supplies and RAID card or other components MTBFs are lower, the latter 
> will dominate the MTBF.
> 
> Take 5 of these units.  I am seeing MTBFs quoted as 10000 to 100000 
> hours for the enclosures.  For laughs, lets take 20000 hours.  There are 
> 8760 hours per year.  5 of these units would consume 43800 operational 
> hours per year.  For a 20000 hour MTBF for these units you should expect 
> a rough failure rate of about 1 enclosure failing every 5 months or so.
> 
> You can ameliorate some of this by building mirror images of these 
> units.  Then you need to worry about the other MTBFs which might not be 
> so well documented.
> 
> Lets for the moment stipulate that the disks themselves are infinitely 
> reliable (they are not, but that is not the point) with zero failure 
> rate.  The other elements of the equation are not as reliable and will 
> fail.  Things like power supplies have MTBFs ranging from 10000 through 
> 100000 hours.  What are the MTBFs of the cables, the USB2 ports, etc? Is 
> there data on this?
> 
> The issue at the end of the day is that what you dont expect is usually 
> what bites your data.  Limiting the maximum damage it can do (N+1 
> supplies, multiple redundant nets, ...) before you can service it is one 
> of your few options.
> 
> Joe
> 

Interesting discussion everyone. My limited experience says given the 
price of redundant but cheap systems and reliable but expensive system, 
one should go for cheapest systems that serves your purpose and 
redundancy than reliable and more expensive system. To elaborate, two 
separate machines with cheap components (chapest SATA drives with single 
power supply) is better that one expensive machine (higher quality hard 
drives, redundant power supply). What you Gurus say?

-Malay


More information about the Bioclusters mailing list