[KLUG Members] Re: RAID

Matty members@kalamazoolinux.org
15 Dec 2002 23:26:32 -0500


Does MD/EVM support mirroring the boot device (/ and company)?


On Sun, 2002-12-15 at 21:14, Bryan J. Smith wrote:
> On Sun, 2002-12-15 at 17:25, Adam Williams wrote:
> > "Should" is an important clause,  I've been bitten hard by lots of
> > "should"s in the past and am very skeptical.  Minor version numbers have
> > often rendered hardware components incompatible and reading the never
> > ending torrent of poorly written docs is just too much effort.  Out of
> > principle I avoid "integrated" components except for the most basic
> > (standard IDE, VGA for a console, etc...) uses.
> 
> Well, if you want to know how I _really_ feel, I think those "BIOS-only"
> ATA RAID cards are complete wastes of money.
> 
> > Sure.  But my only experience is with SCSI.
> 
> Both the Promise SuperTrak and Adaptec 2400A are whoafully underpowered
> compared to most SCSI RAID cards.  They throw in the cheapest,
> lowest-end i960 chips.
> 
> > ATA technology just doesn't interest me.  I find the price difference
> > for reasonable amounts of storage to be negligible,
> 
> "Reasonable" is relative.  "Reasonable" to me is 150-200GB on my home
> workstation, and 200-300GB on my home server.  Of course, I'm an
> engineer and I produce an enormous amount of temporary files.
> 
> > and one avoids all the bizarre versions, hacks, workarounds, and
> > lame products that come with ATA.
> 
> Don't confuse "BIOS-only" ATA RAID with "True hardware" ATA RAID.
> "true" ATA RAID and "true" SCSI RAID are the _same_design_ -- _exactly_!
> 
> As you said, "But my only experience is with SCSI."  I'm going to have
> to insist you are talking from a point of ignorance in that case.
> 
> I've used most brand of SCSI RAID cards with Linux, NT and even OpenVMS
> on both x86 (except VMS) and Alpha.
> 
> > I know what my SCSI card is,  I know it works with 99.999999999% of
> > SCSI drives,
> 
> Depends on your viewpoint.  E.g., 3Ware cards use FPGAs for their ATA
> controllers, so they can be upgraded to newer ATA standards and speeds.
> 
> > no magic cables required, etc....
> 
> Okay, now you're just being argumentative.  If anything, SCSI is the
> _king_ of cable confusion dude.
> 
> Of course, SerialSCSI _is_ being planned as well.
> 
> > Sure you can't get a 10000TB SCSI drive at Best Buy for $19.95 + tax, 
> > but 9 & 18 Gb drives are available cheap in abundance, and 36Gb aren't
> > that bad.
> 
> Cost, cooling, power, etc...  Gets very hard to accommodate.
> 
> Secondly, even 10000rpm SCSI disks of previous-gen technologies can be
> outperformed with latest-gen 5400rpm ATA disks at many operations.
> 
> > An IPS controller can be found for ~$50-$100, less than what some
> > ATA-RAID cards sell for.
> 
> The controller, yes.  But when you need 100GB+, the cost/MB clearly is
> the disk, not the controller.
> 
> > And one can stack a bunch of drives up in an external cabinet with
> > dual P/S and four fans (~$150 on E-Bay).
> 
> First off, trying to buy used hardware doesn't go over well with
> management.
> 
> Secondly, 4 x 80GB in RAID-10 is cheap, easy and fits in most,
> well-designed PC chassis while still being adequately cooled.
> 
> > Drive failures?  I'm still using a cabinet full of 9Gb FH Seagate
> > drives, and they may still be spinning when I'm worm food.
> 
> True.  Many ATA devices are not designed for 24 x 7 operation.  People
> forget that.
> 
> > They don't set any speed records (but you might be surprised) but the
> > majority of the crap on a home system is totally insensitive to drive
> > performance,  just put what matters on fast/new drives (yawningly easy
> > with LVM support) and continue to use all the drives you've already
> > paid for until the croak (still waiting).
> 
> Well, I like to match up models for efficiency.
> 
> > People love to toss around MTBF numbers, etc... but it is my significant
> > cumulative experience that ATA drives are disposable - in large part
> > maybe just because they have to be located in the same case as all the
> > other hot junk - unless one buys extra fans and drive bay coolers
> > and.... so the price differential is even further diminished.
> 
> Yes.  Cooling is important.  Of course, if you purchase a half-way
> descent case with active cooling over the drives, this isn't an issue.
> 
> My workstation (4 x 7200rpm ATA) and server (4 x 5400rpm + 4 x 7200rpm
> ATA) drives stay cool to the touch thanx to an 80mm blowing over them in
> my Antec cases (SX600, 1000 and 1200 cases are used in my home).
> 
> > I'm not interested in starting a SCSI/ATA holy way,
> 
> I think you just did (again).
> 
> The _difference_ between you and I is that I've used _both_, whereas you
> have not.
> 
> > people should use whatever works for them.
> 
> > Just saw my message being interpreted (which
> > is just fine) and thought I'd weigh in with my insight, knowledge,
> > ignorance, experience, prejudice, and general laziness.
> 
> Again, I do have your same attitude in regard to SCSI RAID.  I've
> supported production networks in my career too.
> 
> But I've also made sure I've used _real_ ATA RAID solutions.  And there
> is a heck of a lot of difference between those "BIOS-only" ATA solutions
> and the ASIC+SRAM design in 3Ware products.  Especially for RAID-0, 1
> and 0+1 (aka 10).
> 
> Heck, IMHO, if you're going to go with a microcontroller+DRAM RAID
> solution, it should be SCSI instead of ATA.  Why?  The blocking I/O
> design kills all the performance advantages of ATA anyway, so you'd be
> better off served by SCSI.
> 
> 
> -- 
> Bryan J. Smith, E.I. (BSECE)       Contact Info:  http://thebs.org
> [ http://thebs.org/files/resume/BryanJonSmith_certifications.pdf ]
> ------------------------------------------------------------------
>   The more government chooses for you, the less freedom you have.
> 
-- 
Ryan Matteson - UNIX Administrator
GPG ID: 1B52A210 2002-12-01 Ryan Matteson (Primary Key Pair)
<matty91@bellsouth.net>
Public Key: http://www.daemons.net/~matty/public.asc
Detached Digital Signature: http://www.daemons.net/~matty/public.sig.asc
Fingerprint = A0B1 298E 29C4 3F26 01D5  EDFC 3D62 281F 1B52 A210