Re: RAID: identical disks?

On Tue, 29 Nov 2005 07:08:23 -0500, "J. Clarke"
<jclarke.usenet@xxxxxxxxxxxxxxxx> wrote:

>Curious George wrote:
>
>> On Mon, 28 Nov 2005 11:23:05 -0500, "J. Clarke"
>> <jclarke.usenet@xxxxxxxxxxxxxxxx> wrote:
>>
>>>void@xxxxxxxxxxx wrote:
>>>
>>>> I've read that it's best to set up RAID with identical disks (same brand
>>>> and
>>>> model). I've got a Samsung SP1213N hard drive which I got 21 months
>>>> ago, and
>>>> now I want to set up RAID 1 on my machine. So I'll need to get another
>>>> Samsung SP1213N. But I know that sometimes manufacturers will make
>>>> slight
>>>> changes to hardware and still keep the same model number. I don't know
>>>> if Samsung has done anything to this particular drive in the 21 months
>>>> since I bought mine, so hopefully buying a recent one will be OK to set
>>>> up a RAID mirror.
>>>
>>>In practice, as long as the secondary is of the same capacity or higher
>>>than the primary, and of about the same performance, a RAID1 will work
>>>fine.
>>
>> When disks are mismatched the best case scenario is performance,
>> space, firmware optimizations are limited by the lesser drive.
>
>Yep, but a WD and a Seagate of the same vintage will generally be pretty
>closely matched in performance, space, and firmware optimizations,

Nope. Not firmware optimizations. As far as performance that depends
on what aspect we're talking about as well as whether the two can play
nice.

> and if
>there is a difference in capacity between one bran of 250 GB drive and
>another, it's not so huge a difference that one would consider it
>"limiting" in any but the most pedantic sense.

According to your tolerances.

What's pedantic is to overanalyze a simple statement. I never
inferred an array, esp a software ATA array, would likely be crippled
with mismatched drives. Only that the "lowest common denominator" or
"weakest link", if you will, dictates how the array works.

>> In a
>> worst case scenario it causes compatibility problems. Fortunately on
>> modern hardware esp with software or firmware assisted software raid
>> this worst case scenario is virtually a non-issue.
>
>While that is indeed a worst case, have you ever seen it actually happen?

Yes. Ranging from erroneous PFA failures to massive performance
degradation.

>Most RAID today is software, not hardware,

on the low end.

> and quite honestly Windows and
>Linux and Novell don't _care_ whether a RAID is composed of different
>brands and models

So? In most cases those OS's don't even know what's going on with the
storage on that kind of low level, as they shouldn't.

> as long as the capacity and performance are about the
>same.

nope. They won't care about that either. Esp in the case of firmware
or firmware assisted software raid. The end user might care though.

>>> In
>>>fact there is a school of thought that the two disks should be _different_
>>>brands and/or models on the theory that if they are they same they might
>>>be so closely matched that they both might fail at the same time or close
>>>enough to it that you don't have time to rebuild the mirror.
>>>
>>>Having seen four Japanese-made light bulbs installed at the same time fail
>>>within four hours of each other a year and a half later, I am not going to
>>>denigrate this notion, although I don't consider such failure to be
>>>exceedingly likely.
>>
>> There is most definitely a U-shaped or bathtub curve to hardware
>> failure over time. However both drives in a 2 drive array dying
>> natural deaths within hours of another is quite unlikely.
>
>I believe that I said "exceedingly unlikely" myself. It can happen though.

Nope. you said "I don't consider such failure to be exceedingly
likely." Which since you like being pedantic you should realize it
does not convey exactly the same meaning as "exceedingly unlikely."

>> Rather than mixing models some ppl buy parts from different suppliers
>> to hedge their bets or simply count on premature failure to introduce
>> media of different ages or simply proactively decommission arrays at
>> the end of expected service life rather than wait for the catastrophic
>> event.
>
>Where does one find an "expected service life" rating?

Directly from manufacturers as well as end-user decisions about life
cycle.
.