[nylug-talk] How to set up Raid 1 on Ubuntu, Or Linux in gerneral

[Peter C. Norton]

On Fri, Jun 01, 2007 at 03:28:47PM -0400, sixtyfourbeets wrote:
> On Fri, Jun 01, 2007 at 02:08:24PM -0400, sixtyfourbeets wrote:
> >* If you're buying the same model drive from the same manufacturer of
> *>* the same capacity, then you're getting the same quality with some
> *>* different silicon screwed on to it.
> *>*
> *>* -Peter
> *>*
> *>* Convention dictates that when constructing
> *>* a true SAN one should buy drives from different
> *>* manufacturers.
> *
> No it doesn't. Different drives have different geometries sizes,
> specs, etc.
> ------------------------
> So what?  As long as you buy a drive with the same capacity the SAN/RAID
> hardware/software determines how to place that data on the drive.

Traditionally, not all RAID hardware/software has dealt with widely
different geometries smoothly. Guessing at how to properly decide what
the stripe/block/etc size should be if every disk reports wildly
different physical and behavioural characteristics makes for a poorly
performing array.
 
> ------------------------
> Convention is to get a lot of drives, burn them in, and
> make sure they're under warranty. Real SANs are bought from vendors
> that have done the QA and track the failure rates in the parts they
> provide and they have the relationship to turn around a container full
> of disks and get a new container if they find that N% have failed when
> only Y% should have.
> ------------------------
> 
> 
> >* Doing so addresses
> *>* the concern that a particular drive will not be the same
> *>* as the next drive in the SAN.
> *
> You are contradicting yourself here. Can you clarify this?
> -------------------------
> Putting in HD's from different manufacturers decreases the likelihood
> that multiple drives will fail from a similar defect or when
> it has gone past its MTBF.
> 

Steve, your new convention of using dashed lines in different places
makes it really hard for readers to determine who's writing
what. Please consider a different approach.

> >* The reasoning behind that
> *>* is if drives are from different manufactures chances are
> *>* less likely that the drives will fail in a similar fashion
> *>* in a reasonably short time span.
> *
> It just makes it more likely that you'll have to deal with a greater
> number of different failures.
> ----------------------------
> It usually doesnt matter what kind of failure a drive suffers from to the person
> replacing the drive.

Yes it does. Failures that SMART can detect and predict means that you
have one method of detecting it. Errors from the bus are another
class, another detection mechanism. Failures due to firmware issues
are another class. Failures due to possible connection/hardware
flakiness are another.

Good systems don't just tell you your disk is hoarked, they warn you
so you can pull it before two or more disks go bad at the same time
(like due to lack of warning!).

-Peter

-- 
The 5 year plan:
In five years we'll make up another plan.
Or just re-use this one.