I'm glad we're both calmer now and can come together for real
discussion. I don't expect everyone to always agree with me and
that's fine. This is just one last time to make sure I'm fully
understood. I sometimes wonder if my longer posts are
counter-productive since they're not really geared to the
reading/writing style of Usenet. I also think we got a little more
into word parsing than ideas and drives rather than raid. I'll try to
not to be too long or repetitive. (without promises)
Well I still don't agree but feel this additional post you
made was much more useful than those prior. However a lot
of the things you're promoting are not typically needed for
the environments you're suggesting as good candidates for
the "do it right... SCSI RAID solution".
My _opinion_ is that if you don't NEED _extremely_ high availability
or _absolute_best_ data integrity protection you probably don't need
RAID. If you do NEED RAID than it should be a "top-notch"
implementation or not to bother. I say this primarily because just
because a card or software generates redundant data it isn't
necessarily a foolproof fail-safe that will end up saving you time &
effort down the line or increase performance. Granted I could
probably take "ability to flash upgrade without reboot" & "redundant
dedicated fail-over or cache-mirroring paths" off the list - but I'm
not sure I would take much more off even though its for home or SOHO
use. I guess its fair to expect some disagreement on this. I will
characterize Linux, Novell, and SNMP support as 'optional' - depending
on the home network.
Partial explanation/support:
RAID requires a decent chunk of time doing research about it &
products, experimenting with stripe size, levels and other
configuration options, observing behavior/speed, testing recovery
scenarios/strategy (& that's in the best case scenario without
compatibility problems or discovering bugs). Overall productivity &
free time goes down significantly if you spend a lot of time setting
up and administering something you totally don't need or as insurance
for an event that may never happen during a short service life, or
which may be ill-prepared for certain failures you are unlucky enough
to have - all while reducing overall storage MTBF characterization.
The RAID must therefore deliver a very significant timesavings & data
protection benefit to offset this initial time, $$ investment (incl
parts, additional electricity, building in a better box, etc.) and
risk from increased complexity. No product can be seen as "reliable"
if it has difficulty or potential difficulty meeting its core
purposes/promises (in this case it is a data
security/reliability/availability/performance 'boosting' product).
Because the standard of function that must be met in order for it to
"keep all it's promises" is so high there is much less of a problem of
something being overkill than insufficient. Being able to get a bunch
of drive to work together and generate ECC data is not really the
whole poop.
Don't get me wrong RAID can be fine to use anywhere - even in home -
if you are comfortable with it and have very valuable/limited time
relative to data quantity so it has enough potential to be of real
help. But I just can't make that same big distinction between few
users/home & many users/business for general "reliability". For the
most part availability is availability and for the whole part data
integrity is data integrity regardless. You may have less users
relying on the data but you also have less human resources to manage
and secure it. Time is just as valuable when you're home. It's
probably more valuable because there is so little of it. (and yet I
keep typing...)
Call me paranoid but I'm also always suspicious of products with big
promises that try to lull me into a sense of security esp when there
are huge price discrepancies (& I mean suspicion across the whole
price range).
No immature technology is a good choice
True.
and yet there is no assurance
that any particular, specific SCSI controller tech is more
mature than an ATA..
No assurance- well OK. That's why I've been qualifying "better scsi"
or the "best scsi products" instead of claiming "all".
"Mature" is a tricky word 'cause it implies two things.
1. Track record: With many scsi product lines it's hard to argue
"track record" because the companies got bought out so often so the
product lines are interrupted. With others it is easy and they win
hands down.
2. In terms of "robustness" most scsi beat all ata hands down until
rather recently. Now its more case by case- except at the top tier
(like some reputable san stuff, etc) which blows away the best SATA
hands down.
Now there is currently a place for SATA RAID in the enterprise, but
it's mainly near-line storage, caching for tape libraries, etc. It's
a hard sell for more important roles partly for performance and partly
for not yet being "tried & true." Robustness is generally hard to
convey and compare for a client who isn't already confident in a
"track record".
If a company (like LSI for example) is simply migrating the same
technology from SCSI to ATA it is the same & just as "mature" (as far
as "robustness" but not "track record")- provided, of course that the
entire feature set is ported and there are no kinks in the process
that haven't yet been worked out or cause them to make very large
revisions/redesigns.
Implementation of RAID levels is for the most part proprietary; it
isn't entirely standardized. So I still think different offerings
merit close scrutiny esp. from companies who haven't done this type of
thing before for 'enterprise' use- so their ATA raid design goal from
day one could very well be to sell cheap, sexy storage to
'enthusiasts' and they therefore feel different customer obligations
and pressures. (I can hear the flame being typed now) There also is
one or two mutant raid levels only available on ata which makes me
weary of their claims of "robustness" as there is basically is no
track record and its a hard comparison. If they are indeed bad "disk
quality" will have nothing to do with array "reliability" in those
cases.
You may not agree but for personal storage, apart from certain
performance and design differences, quality control is relaxed because
of the relative tradeoff in profitability / defect rates. For ATA
drive manufacturing the percentile component rejection rate is
generally around 5x less rigorous than scsi drives. Since ATA drives
ship at a rate of around 6 to 1 over scsi, that amounts to a huge
difference in total questionable units. I don't know which components
tend to have higher failure rates and which supply sources are less
reputable and how that fits to individual lines. Trying to pin this
down more specifically without discussing 'inside information' is not
really possible. I can only legitimately talk about anecdotal
experience about operational success- which is what I meant early in
the thread when I talked about experiential vs engineering info in
relation to you contention about drive quality & reliability.
Of course independent reliability surveys put both kinds of devices
all over the map. That's why I didn't say all scsi drives are more
reliable than all ata ones and tried to focus on a line I've had good
experience with which is corroborated by "independent" parties (to try
to synthesize a somewhat representative sample).
This 'quality' difference makes sense for many pro vs consumer
products because with larger profit margins you can afford to tighten
quality control as well as employ better programmers and engineers and
devote more resources to testing & development, etc. In addition you
have more savvy customers who demand more from the products who you
have to satisfy with more conservative products from a reliability
standpoint. You're right, though, that there is no _assurance_ that
companies will always "do the right thing" for their consumers and the
"top of the line" is often exploitative of consumers with deep
pockets. Furthermore less units are produced so the actual difference
in profit margin isn't exactly what it outwardly appears.
Believe me though you don't want a bad _batch_ of drives in an array.
Having drives that are less susceptible to dings and vibrations and
quieter (generally FDB over rotary voice coil) more geared to heavy
constant use, and with advanced features to preserve data integrity
(like background defect scanning), and with flexibility to make the
raid work better/more compatible like upgradeable firmware and mode
page settings all come together to make a more robust "do it right"
type solution with scsi drives. (of course you first have to agree
with my cost/benefit overview to see the necessity for this & specific
recommendations should indeed change as available product attributes
change)
but it was never meant to be a direct comparison between the two
But we have to compare the two in order to settle your objection to a
good scsi RAID 1 (or variants) using a well regarded scsi drive family
is a safer bet towards the reliability end over ATA offerings as well
as whether ATA and SCSI drive quality & reliability are equal
"there isn't some unique failure point or 'lower quality' part on
ATA drives that makes them more susceptible to failure."
As far as the initial recommendation - I could have also included RAID
3 & 4 but it requires better HW. I didn't mention exact cards because
with the price it kinda depends on whether what falls in his lap is
acceptable. I also wanted to slow him down because a "newbie" looking
for cheap raid 5 is in for surprises as he learns more about raid.
"reliability" is NOT the same thing as features.
Yes in terms of semantics. No in terms of the idea I'm trying to
convey.
We're going to have to agree to disagree on this point. If you have
to fiddle with it or take it down once in a while it is not "reliable"
because it is not meeting its main purpose of very high availability.
Likewise without a full feature set that ensures data integrity it
would not be "reliable" when or if corruption is generated during disk
failure, power failure, flaky devices, noise, or whatever the specific
vulnerabilitie(s). It also isn't "reliable" if you are operating with
a controller problem & there are no diagnostics to pick it up (I have
personal experience with diagnostics mitigating loss) or with a bad
configuration backup/restore features that interfere with getting an
array back on-line - (availability expectations/time expense). You
get my point - I won't go through every line item.
The main problem is that array "reliability" is inversely related to #
spindles and also by the same calculation lower than a single disk.
These "features" that combat _all_ the reliability concerns are
vitally important for raid to bring real benefit over a single disk
and enough benefit to justify _all_ inherent costs.
I've already posted that I was done with this thread but
posted again only to compliment you on taking the time to
more cleary express your concerns with the differences...
regardless of whether I happen to agree with them in the
context used.
Thank you for taking the time to read my post and comment on it
putting aside our harsh disagreement. I hope we can continue this
tone in future threads. This sounds bizarre but I'm glad we still
don't agree. We all learn in these forums by presenting & hearing
different views - so long as they are forthright and explained. (yeah
I know that sounds like insincere cheesy BS but it's not really)
Part of why I kept responding was that confronting misconceptions
about raid as well as exploring what "reliability" means is of benefit
to the group. Another was purely selfish, as I was really hoping to
force out compelling evidence that large cheap SATA raid is "proven"
and ready to replace some other more expensive installations. I'm not
saying this to keep contention alive. In fact you made me re-look at
and reconsider product lines and a few newer products seem much more
compelling than ones I saw available even a few months ago. I still
think though I'm going to wait at least another product cycle or two
before putting a new sata in the test lab again in plans for use and
recommendation. But the day is drawing closer...
I hope we haven't scared the group off continuing to discuss more
detail of raid in future threads. I may have overdone it here. I bet
everyone is sorry you pressed me so hard for "details" (if there is
anyone still reading).
