RAID-1 reliability

[dorothy.bradbury]

No it is not bad advice to claim RAID-5 is more reliable than RAID-1
o You are citing *theoretical* comparison
o I am citing *practical product offering*

Yes:
o RAID-1 uses fewer disks than RAID-5 so fewer disks to fail
o So the probability of a disk failure in the population is less.

However, the issue isn't theory, it's practice & product offering:
o RAID-1
---- usually IDE based -- 1yr warranty & past design problems
---- usually complex recovery -- for Windows s/w RAID or promise RAID
---- UPS & closedown required for write-cache flushing on many
o RAID-5
---- usually SCSI based -- 3-5yr warranty & less history of problems
---- usually auto-rebuild & hot standby -- so recovery is fast & seamless
---- usually battery backup of the write cache as standard

YES:
o There is RAID-5 from Adaptec & 3ware for IDE.
o The operating system is a major factor.
o There are h/w transparent RAID-1 systems with hot-standby & auto-rebuild

However for reliable business use I would:
o Go for RAID-5 by Mylex or similar SCSI
---- RAID-5 by 3ware on IDE if data-set large re SCSI drive cost
o Over RAID-1 by low-end IDE
---- since it is offers a promise of

You can set a Mylex controller to RAID-1 if you wish.
For large arrays, yes, RAID-5 does become problematic in that you can lose
multiple drives at once - cooling failure or poor enclosure design, very
easy.

 

[J.Clarke]

You seem very confused. RAID 1 vs RAID 5 is a different issue from
cheap vs good.

RAID-1 is a redundant storage configuration consisting of two drives
containing identical data. RAID-5 is a redundant storage configuration
consisting of 3 or more drives with data and parity distributed across
them.

RAID-1 can be implemented with SCSI, IDE, Fibre Channel, ST-506, or
any other interface technology, using dedicated hardware or using
software.

RAID-5 can be implemented with SCSI, IDE, Fibre Channel, ST-506, or
any other interface technology, using dedicated hardware or using
software.

IDE is a drive interface. It can be used to implement RAID 1, RAID 5,
RAID 0, RAID 3, or any other RAID level.

SCSI is a drive interface. It can be used to implement RAID 1, RAID 5,
RAID 0, RAID 3, or any other RAID level.

In general, any dedicated hardware that supports RAID 5 will also
support RAID 1.

You don't seem to be able to separate the configuration from the quality
of the hardware.

No it is not bad advice to claim RAID-5 is more reliable than RAID-1
o You are citing *theoretical* comparison
o I am citing *practical product offering*

Fine, please demonstrate that using the same hardware and operating
system RAID 5 will provide greater reliability than RAID 1.

And I did not say anything about "good advice" or "bad advice". I
simply pointed out that your statement was incorrect, which it is. RAID
5 is adequate for most situations in which redundant storage is
required. But it is not the most reliable alternative, only the most
cost-effective. The "most reliable" is usually not needed.

Yes:
o RAID-1 uses fewer disks than RAID-5 so fewer disks to fail
o So the probability of a disk failure in the population is less.

However, the issue isn't theory, it's practice & product offering:
o RAID-1
---- usually IDE based -- 1yr warranty & past design problems

There is nothing that requires that RAID 1 be IDE based. "Past design
problems" are past--do you know of any _current_ design problems that
are inherent in RAID 1 and not in particular devices which are sometimes
used to support that RAID level? Leaving aside the issue that RAID 1
does not have to be IDE based, there are, you know, IDE drives available
that have 3 year warranties and that use exactly the same mechanical
components as enterprise-level SCSI drives.

---- usually complex recovery -- for Windows s/w RAID or promise RAID

The recovery for Windows with s/w RAID 5 is also complex. So what?
The issue there is software vs hardware RAID, not RAID 1 vs RAID 5 or
even IDE vs SCSI.

---- UPS & closedown required for write-cache flushing on many

The same is true for most RAID 5 systems in existence, including the
majority of deployed SCSI systems.

In any case, I am curious--are you advocating that mission-critical
servers be run without UPS protection and automated shutdown?

o RAID-5
---- usually SCSI based -- 3-5yr warranty & less history of problems

There is nothing that requires that RAID 5 be SCSI-based. It may also
be implemented with IDE or other interfaces. You are again confused on
the issue of RAID level vs hardware quality.

---- usually auto-rebuild & hot standby -- so recovery is fast &
seamless---- usually battery backup of the write cache as standard

Please demonstrate that "usually battery backup of the write cache is
standard". It has been extra cost on every RAID controller that I have
ever seen that supports that feature. But again that is a red herring
that has nothing whatsoever to do with RAID 1 vs RAID 5, instead being
an issue of high end vs low end hardware.

You can use a crappy RAID 1 setup and have problems. You can also use a
crappy RAID 5 setup and have problems. You seem to be saying that RAID
1 is necessarily implemented with crappy hardware and RAID 5 is
necessarily implemented with high quality hardware. Neither is the
case.

YES:
o There is RAID-5 from Adaptec & 3ware for IDE.

And from a number of other vendors, including Promise.

o The operating system is a major factor.

Only to the extent that there are or are not drivers for a particular
OS.

o There are h/w transparent RAID-1 systems with hot-standby &
auto-rebuild

Almost all h/w transparent (that term makes no sense--perhaps you
meant s/w transparent?) RAID-5 capable systems which have those
features are also capable of RAID 1. There are also RAID-5 systems that
do not have those features.

However for reliable business use I would:
o Go for RAID-5 by Mylex or similar SCSI
---- RAID-5 by 3ware on IDE if data-set large re SCSI drive cost
o Over RAID-1 by low-end IDE
---- since it is offers a promise of

You can set a Mylex controller to RAID-1 if you wish.

Yes, you can, which is precisely my point. RAID 1 vs RAID 5 is not a
hardware issue, it is a configuration issue.

 

[Arno Wagner]

No it is not bad advice to claim RAID-5 is more reliable than RAID-1
o You are citing *theoretical* comparison
o I am citing *practical product offering*

You are comparing an expensive RAID-5 solution with a very cheap RAID-1
solution. That is not fair, as there are more expensive RAID-1
solutions. True, it is more difficult to get a cheap RAID-5
solution, so if you take average quality of RAID-1 and RAID-5
solutions over wehat is available on the market you are certainly
correct.

But if you take RAID-1 and RAID-5 from the same quality/cost class,
RAID-1 is massively more reliable any time, since my figures are
quite applicable in practice.

Yes:
o RAID-1 uses fewer disks than RAID-5 so fewer disks to fail
o So the probability of a disk failure in the population is less.

However, the issue isn't theory, it's practice & product offering:
o RAID-1
---- usually IDE based -- 1yr warranty & past design problems
---- usually complex recovery -- for Windows s/w RAID or promise RAID
---- UPS & closedown required for write-cache flushing on many
o RAID-5
---- usually SCSI based -- 3-5yr warranty & less history of problems
---- usually auto-rebuild & hot standby -- so recovery is fast & seamless
---- usually battery backup of the write cache as standard

True. And usually (with this) cost(RAID-5) >= 10*cost(RAID-1).

O.k., I think we are both right. Our approch is just different.
Mine is: Balance cost of loosing data against risk of the solution
coosen to fail and cost of the solution.

Yours is: Select a solution from the market that fits the need
and is simple to administrate and then get the funding for it.

If the infrastructure should be kept simple, your solution
is better. If cost should be optimal, my solution could
be better. Since I do not factor in maintenance and
administration cost, while you do, we are really targetting
different scenarios. Depending on the setting, the lower
maintenance cost and risk of selecting the wrong solution
can compensate the higer hardware cost. But it will not
always.

YES:
o There is RAID-5 from Adaptec & 3ware for IDE.
o The operating system is a major factor.
o There are h/w transparent RAID-1 systems with hot-standby & auto-rebuild

However for reliable business use I would:
o Go for RAID-5 by Mylex or similar SCSI
---- RAID-5 by 3ware on IDE if data-set large re SCSI drive cost
o Over RAID-1 by low-end IDE
---- since it is offers a promise of

You can set a Mylex controller to RAID-1 if you wish.

Yes, that is very advisable if capacity of one disk is
enough and writing speed is not critical.

For large arrays, yes, RAID-5 does become problematic in that you
can lose multiple drives at once - cooling failure or poor enclosure
design, very easy.

Personally I start to feel unfomfortable when there are 5 or more
active disks in a RAID-5. That is why I operate several RAID-5 with
3/4 disks at the moment. On the other hand these are IDE disks
in software-RAID, so cost of the redundcancy is low.

Still, for people that do not really know about storage
technology, need reasonable reliablility and have the cash
to spend, RAID-5 with SCSI is probably the best and easiest to
use solution.

Arno

 

[Rod Speed]

message
Some of this stuff is mindlessly superficial.

o RAID-1
---- usually IDE based -- 1yr warranty & past design problems
o RAID-5
---- usually SCSI based -- 3-5yr warranty & less history of problems

There's plenty of 3 years warranty IDE drives around which
dont have 'past design problems' and its mindlessly superficial
to claim that SCSI in general has 'less history of problems'

And you keep ignoring the other fundamental that when
IDE drives are so much cheaper, you can end up well
ahead when you spend the same amount on the IDE
based system and do something sensible as far as
spreading the drives over more than one PC and end
up with rather better reliability for the same total price.

 

[dorothy.bradbury]

something sensible as far as spreading the drives over more

than one PC and end up with rather better reliability for the
same total price.

So the saving of IDE now pays for duplicate motherboards,
processors and case? Remarkable. Whilst 10k-rpm drives have
at least made the light of day, SCSI drives offer performance.
Spindle speed is a major factor in latency and 15k is SCSI only.

I'm also happy to argue SCSI drivers offer reliability over IDE.

The reality is probably Serial-ATA offers fresh pricing power
for manufacturers to extend warranty at minimal cost to them.

For IDE RAID, I'd want 0+1, ie, 10, rather than 5.
Serial-ATA breaks the disks/channel problem/cost of IDE.

Conversely, where data-set capacity is low, then SCSI is a
lot more attractive in cost and controllers are quite economic.
Sure 3ware IDE are good - but their prices aren't low either.

 

[Rod Speed]

Yes - in terms of being fobbed off with some of the "it's RAID"
IDE solutions out there which are often pushed by integrators.

Or RAID1 can have some advantages over no redundancy at
all, and can be useful when the expensive end of redundancy
cant be justified because the need for redundancy is mostly
just for convenience, not when uptime is crucial.

I'd add, though, that it's application dependent:
o RAID-5 for a database server can be less than ideal
o Big array, lose many disks = array rebuilding may not be possible
---- same applies to RAID-1 if both source & mirror fail
o Big array, lose a disk = rebuild time can be very long
---- recovery possible, but responsiveness can be impacted

Classic example of the latter was an ISP with NNTP server failure:
o Rebuild from a (big) RAID crash took hours
o Rebuild from peering servers took hours

A spanner comes from RAID 0+1 / 10, possible on cheap IDE.
It avoids some short-comings of RAID-5 & has many benefits.

However, on the cheap IDE 0+1 solutions, the question again is
how easy is recovery, how long & what human error risk to data.

It can be useful when compared with the alternative of no RAID at all.

With full backups as well.

I like RAID 0+1 for web dbase servers, can be done cheap.

And reduces the urgency when a drive fails.

That can divert money into a failover server or better coloco facility.
Linux too can help here, but that's wandering off topic (as usual

Yep, has little relevance to what was originally being discussed,
whether using identical drive models with a low cost RAID1
approach increases the risk of losing both drives at once.

It doesnt.

 

[Rod Speed]

So the saving of IDE now pays for duplicate
motherboards, processors and case?

It can do, most obviously when there
is more than one PC there already.

Remarkable.

Basic common sense, actually.

Whilst 10k-rpm drives have at least made the
light of day, SCSI drives offer performance.

And those choosing to use cheap RAID1 and full backups
just to minimise the hassle of a single drive failure when
modern high performance IDE drives are now so cheap,
dont need that performance with personal desktop systems.

RAID1 can be a cheap way of avoiding a major hassle with
a single drive failure. Its easy enough to handle many of the
other sources of failure, just plug in another keyboard or
monitor etc, but a full restore to the replacement boot drive
will always be more hassle than the use of RAID1 and some
may choose to pay bugger all for the extra convenience.

Spindle speed is a major factor in latency and 15k is SCSI only.

And few need lowest latencys with personal desktop systems.

I'm also happy to argue SCSI drivers offer reliability over IDE.

Easy to claim. Have fun substantiating
that claim with reliable statistics.

And the whole point of RAID1 is that single
drive failure is just a nuisance if it does happen.

Can make a lot more sense to have a pair of cheap IDE
drives with 3 year warrantys in a RAID1 config than to
pay thru the nose for the purported higher reliability of
say a RAID5 SCSI array of the same total capacity.

The reality is probably Serial-ATA offers fresh pricing power
for manufacturers to extend warranty at minimal cost to them.

In reality its just a change thats become necessary as the
drive performance starts to stretch what PATA is viable for.

For IDE RAID, I'd want 0+1, ie, 10, rather than 5.

Sure, there are certainly some advantages with that approach.

Serial-ATA breaks the disks/channel problem/cost of IDE.

Nope, PATA can be used like that too.

Conversely, where data-set capacity is low,
then SCSI is a lot more attractive in cost

Its always more expensive than using mass market commodity drives.

and controllers are quite economic.

Nothing like as economic as whats included on the motherboard.

Sure 3ware IDE are good - but their prices aren't low either.

The drives are tho.