I recently had a motherboard mounted Promise RAID controller tell me a disk “has failed or been disconnected” in my RAID 1 array. I tried several reboots over two days but always the same message. Not to worry, that’s what RAID 1 is for.

I ordered another disk and when it arrived I booted once more to check the disk number to replace and the damn thing recognised the disk again. So I rebuilt the re-recognised disk from the good disk, getting everything back as before. I know, this is not a disaster, but what about the risk of running on 1 disk for a while and the spend of money and time?

Anyway this left me wondering about the reliability/sensitivity of the controller. I decided it was probably still a disk issue, so I changed it anyway, just in case; the new disk is bigger and quicker anyway. But still I do not know the real cause, no errors were reported when I checked the recovered disk.

Later I thought about the whole thing again and decided that in future I will change disks about every three years regardless. The advantages of a scheduled replacement with no time on one disk and a performance boost to the slowest item on the computer far exceed the little money I would save by trying to squeeze the last drop of life out of a disk.

I have had very wide variation in reliability of disks, from less than a year to well past 4 years of hard use and recently came across one with over 7 years on its clock. By then a modern graphics card typically has more RAM that a disk that old has free space. I also have to say in defense of RAID 1 that it has saved me the grief of a system rebuild three times now and I would sooner it be too keen to criticize a disk than too slow. I think all computers should be sold with a hot swapping RAID 1. In fact I marvel at how cavalier people are running without it. Sure backup can get you out of a hole but why get yourself into it in the first place.

I have an old HP server that hadn't been powered down in at least a year. When it had to be shut down one of the drives in the RAID-5 array just would not spin back up.

Got a replacement drive, the array rebuilt automatically and everything has been fine since. (Of course it probably hasn't been powered down since :)

HP SmartArray for the win. The external arrays are self configures, and thus swapable from server to server, which is very nice when a server takes a nose dive without notice. We lost a file server recently from lack of defraging (From lack of space :/). This server was accessed constantly, by 100 people from 7 AM, to 9 PM, for six years, with no more than a reboot every three months. Finally disk queue times brought it down.

I've never not been able to save a hardware, non-raid 0 array from a drive failure. Software RAID however, I have yet to get to work the way it should. The 2003 server built in tends to have dissappearing reappearing hard drives if you convert them to Dynamic (Other users can't see/edit them)

I've had individual drive failures on an Infrant (now Netgear) ReadyNAS NV that have taken the whole box offline. Yeah, I know it's not _supposed_ to happen that way. Newer firmware doesn't seem to have some of the same issues, but they really need a complete firmware rewrite from the ground up, and Netgear doesn't seem to want to put these kinds of resources into the product.

I've heard of catastrophic RAID failures caused by replacing a failed drive, but never experienced anything like that myself. Apparently drives can develop bad spots that are unused, and can encounter them when rebuilding an array, and the resulting failure can take out the degraded array. Some RAID systems periodically scrub (read then rewrite) all the sectors on all the disks to prevent this kind of problem, but I don't know of any SOHO products that do. Yanking and replacing a drive to force a rebuild thus isn't a horrible idea, though it will degrade the array and/or slow the device down for a while.

Since you really gotta have offsite backups for your critical data, failure of a RAID array shouldn't be fatal, I use nightly rsync backups to an external USB drive to preserve my critical data, and use Sneakernet to occasionally copy that data to another NAS drive in another location. Not perfect, but not awful either.

Murphy says that you are going to have {drive, array, power supply, network, fire, flood, acts-of-god} failures, so you should have multiple layers of redundancy in your backup process.

Last year, the RAID controller in our 2 month old Dell PowerEdge 2950 suddenly died, killing our RAID 5 array with it. We got the machine replaced, and restored from backup the next day, but realized that RAID 5 gave us a false sense of security in that the controller was a single point of failure.

Acting on a dumb idea, I decided to switch my home PC's HD over to a RAID 1 drive system some months ago. I got one of those Sil3124 cards (Addonics), did some snazzy WinXP fiddling, and felt pretty safe knowing that if one drive failed, my PC would continue humming along.

However!

About 2 weeks into this new hardware/software configuration, I woke up to a BSOD. After some detective work, I discovered that the RAID 1 had dropped offline. (How does a directly-attached RAID 1 drop offline?!) After a bios-induced rebuild, I was back up and running. Thankfully one of the two drives is always "clean", so all I need is a rebuild. But every 2 or so weeks it still hiccups and dies, and I have to rebuild it. After a few months of this, I ask Addonics what they thought - no reply at all. I'm now considering going back to my single-drive configuration, with nightly backups onto a RAID-driven home NAS. I swear I've never had so much trouble with my PC's HD than when I switched over to using a RAID 1 array for it!

(I have since been laughed at...er...informed by many IT people that the Sil3124 is absolute crap.)

Whenever I'm dealing with a new server (or any computer/RAID set up for that matter) which I'm not familiar with, one of the first things I do is test the basics of the RAID controller and simulate a drive failure. As some others have said, RAID can sometimes give you a false sense of security. Essentially, I'll take out a drive and see if the system even boots, and then find out exactly what is necessary to get the system back up and running properly.

Some RAID controllers are exceptionally easy to deal with, but others can be a royal pain. I dealt with an HP ML-310 once that had a failed drive and was extremely picky as to how it would rebuild the RAID. Some of my first experiences were on some older Dell Perc controllers where my *assumption* was all you had to do was pop in a spare drive and then sunshine and Roses appear.

Common sense I figure, but absolutely necessary. And there's always something to be said about when the controller itself dies. Typically, with name brand servers you can rest assured HP/Dell will send you the exact replacement, but short of those scenarios getting back up and running can also prove painful.

A lot of newer SAN's/Servers come with two, and three controller cards to avoid this, however those cost extra money as well. The SAN's we use are even called "Zero Point Failure" SAN's. When one controller goes out, another takes over often times stream lined. For servers it will however take a reboot (For the older non-hot swap). These days everything in a server is "Hot Swappable". We use Massive HP DL-580's that take ages to start.. so restarts for hardware failures are a no go.

Re: I always test recoverability:

That is by far one of the best practices for anyone, before you put a server into production, make SURE you know how to recover it ASAP.

ML-310's, yack. I use an old ML-370 in my house (Makes a good space heater if nothing else!). The biggest bane of any older HP servers, to edit the RAID controller settings, you have to physically boot the server with the SmartArray CD. This still boggles me to this day. The newer G5's, and I think G4's can be edited inside windows I know, however I'm not sure about other OS's.

It's kind of funny when you think about it. You can take a raid 50 array (2 raid 5 arrays put into a raid 0). and have amazing performance.. however as performance goes up, data security goes down. The only RAID that offers fool proof data protection is a mirrored RAID with three or more drives.

In the end, RAID really isn't for protection as much as performance, no matter how many points of failure you take a way, you're still taking a chance at losing everything.

Ive had 0,and 1 raids, both have failed because 1 wd raptor failed/crashed...
couldnt rebuild the 1, for "unknown parity error? using promise onboard ASUS mobo

I try to avoid motherboard RAID controllers. It's like joining the military, the cheapest bidder makes the equipment that keeps you alive.

I've never used ASUS desktop motherboards, however I know the server boards (At least the newer ones) use Lsi Logic controllers, which can more than hold their own(LSI that is, I’ve only used a few of these boards).

This isn't saying all are bad, I would just be careful, more so when using expensive hard drives like WD Raptors

I have a couple of failures on the log. All recovered without problems. A few times disk failure in dells raid5 array and a controller failure in a hp server.

Bought 1TB MSSII after reading good review here. Setup for RAID 1 mirroring for maximum protection. RAID failure per blink codes, all NAS data lost but had backup USB drive. Replaced via Seagate $20 2nd-day air warranty service with refurbished drive. Restored data from backup USB drive. Refurbished MSSII drive failure after 3 months (data intact this time). No longer under warranty. Replaced with an open BYOB Synology NAS. So far so good.

My Raid 0 failed (Maxtor HD) twice,First time I did not backup any data so it was a total loss.Second time Promise FastTrack TX2000 raid controller failed (Raid 0) but this time I did have a backup.
Now I build the array again but this time using WD Raptors 10K,and Raid 0 1.But for sure I don't trust Raid any more,you get fast system better read write speed then single disk but no reliability.

I HAD AN OLD NAS WITH A DEAD DRIVE IN RAID 1 AND THE ONE THAT WAS LEFT WAS ON ITS WAY OUT. I PURCHASED A NEW D-LINK DNS-323 AND TWO 750GB SEAGATE 7200.11 32MB DRIVES. IT WAS AN EASY SETUP AND IT SEEMED TO BE FINE SO I MOVED ALL MY DATA TO THE NEW DRIVE. I WAS JUST FINISHED WHEN I SAW A YELLOW LEFT HARD DRIVE LIGHT. I TRIED TO ACCESS THE DRIVE AND NO LUCK. I CALLED TECH SUPORT NO HELP! MY DATA WAS GONE. NEEDLESS TO SAY THE DNS-323 WENT BACK AND I FOUND YET AGAIN YOU GET WHAT YOU PAY FOR SO I PICKED UP A READYNAS NV W/2 750GB HARD DRIVE AND WORKED IN TWO MORE 750GB. IF I HAD NOT RUN INTO THIS SAME TYPE OF PROBLEM IN THE PAST I WOULD NOT HAVE HAD A BACKUP ON USB HARD DRIVES. ONCE RESTORED I HAVE BEEN UP AND RUNING FOR OVER 6 MONTHS. I STILL KEEP A CURRENT COPY ON A SECOND MEDIA. ANYTHING WE PLUG IN CAN (AND WILL) BREAKDOWN OVER TIME SOFTWARE RAID OR SOMETHING ELSE BACKUP YOUR BACKUP.

I used the onboard raid controller from my asus motherboard in raid1 mode and launched an antivirus scan one morning. A couple of hours later when I got back I was greeted by a black screen which wouldn't wake up. After rebooting I got errors on the array.
Some fiddling with everything later got me to the conclusion that both drives had failed on the same day.

the conclusion I took from this incident:
* never launch antivirus scans on a hot day :)
* raid is good (because both disks didn't fail on the same spot I was able to recuperate almost everything)
* keeping a backup for critical files is even better

The one thing I have never seen from websites like yours and only from some others are reliability measurements. Probably because it costs a lot of money to house the equipment and much of it is on loan to you as a reviewer.

Dynamix, which was part of the now defunct Sierra Software, had a group of people like myself that did beta testing for Tribes and Starsiege. We were called the "Players Council". We were under NDA and got beta builds of the software. I don't see why we can't do the same thing and send you data about our infrastructure (IBM and HP hardware) as well as getting some stuff for home as a long-term reviewer. I don't have a lot of time to do intense short term reviews and write ups but I can keep a daily log of things and of course torture them.

Think about it.

Also, SATA RAID cards (notice I didn't say SAS?) and older IDE RAID cards were software only for the most part. A lot of the logic depended on software drivers continuing to work properly for the rebuild of RAID cards. Intuition will tell you that this isn't robust. They just aren't good solutions. Nice marketing ploy to sell add-in IDE controllers though.

I have a lot of consumer level stuff we try to wedge in here for cost
eturn reasons. Small offices in China and Africa insure that mobility is key. I can give you some pretty nasty stats on the Netgear (previously Infrant) ReadyNAS and the torture it can be put through when on 3rd world country power. (the rebuilds are freakin slow though). I will email you my response address if you want some information on this and other solutions (Cisco 871 with POE add-in, Linksys SD208, and I'm currently beta testing the new Vonage Analog Phone adapter w/ LCD display). Cheers

For Windows, 3ware RAID 1 has been rock solid reliable. The optional battery backup unit (BBU) has saved me several times when Windows has crashed or the power goes out for a long time (exhausting the UPS battery). Sometimes Windows does not shut down property via the UPS shutdown and it is great to not get your data messed up. Instead, you get to power back up and see the 3ware log message of the journal being written out to the disk. Awesome. I've rebuilt a few times from drive failures and it has worked fine every time. The worst part of a drive failure is dealing with the shoddy RMA service that Western Digital and Seagate provide.

Software RAID/Host RAID under Windows is mostly just asking for problems. I have learned this the hard way and find a 3ware card a very worthwhile investment. Even my two Linux boxes utilize 3ware cards. They just work.

Intel's motherboard raid (Matrix Storage Manager) forces you to rebuild the set after the bios is reset. You don't lose any data, but it takes forever to rebuild.

I was my problems came from my drive selection. I bought four best bang-for the-buck Western Digital Caviar SE16 WD5000KS 500GB 7200 RPM SATA Hard Drives and wanted to put them in a RAID 5. Seemed to work OK, but every third or fourth week, one of the drives fell out of the array. You can imagine how happy I was to find the system with two dropped drives. Mind you... these drives aren't bad, just not happy to work in a RAID. Through magic, I was able to recover the data. And set the system up as 4 independent volumes.

I knew a firmware patch went out for the WD5000YS to fix some raid issues (http://www.theinquirer.net/gb/inquirer/news/2007/01/25/western-digital-releases-fix-for-vanishing-caviar-drives), but never my model. I really like the idea of RAID. Guess if I ever do it again, I don't assume "drives are just drives", even though RAID 5 is a replace and be happy system. Never imagined that there were "RAID-Unhappy" drives. I now worry about how wide of an issue this really is!

Most of my experience has been with linux software raids. I've recovered one raid (reason it went down was due to repeated dirty shutdowns (due to bad memory)). Pretty painless, if you know what you're doing.

I've worked with a server that had scsi raid through an adaptec controller. The power supply died (took the motherboard on its way out, but left the drives usable), and I transplanted the whole OS installation (disk subsystem and all) unto a similar hal. Booted right up and picked up where it left off.

It was during the research for that I came across that a lot of those controllers use propitiatory encoding. (Is this still the case?) So since that experience, I choose linux software raid to avoid the propitiatory encoding and recovery headache that goes with it. You have to use a transport layer like NFS or CIFS to get at it, and therefore it's sometimes slower.

Haven't lost too much yet. I know you're supposed to use identical hardware, but I always stagger the drives (age, manufacturer, supplier), lest the day come when they all decide to hit their expiration.

I started using SpinRite on all of my New HD's in Maintance Mode before I put them in service. This will locates all of the bad sectors that mfg lean on the SMART Tech to catch and mark out of service. What happen when you have 2 consective bad sectors (quite common), controllers see this delay as a bad drive, and fails the array. Running SpinRite eliminates these errors and you get better performance. I have used SpinRite to repair HD's that was reported as bad. I only use the Raid Edition drives now, I have found that they cause less problems in the long run.

The Smart technology is nice in detecting failing drives, but can raise havic on raid arrays.

Strangely, you can't always simulate drive failures, especially by merely yanking a drive. My Infrant/Netgear ReadyNAS _hung_ when a drive went bad (because the drive didn't respond properly?), but handles the "unexpected drive removal" without a problem. [Putting the drive back in in RAID-1 mode doesn't work, but that's a whole nother rathole.]

Haven't have RAID failiure but we have 3 readynas 1100's that have lost a share and it cannot be recovered without remote intervention, which for us is impossible, for what these cost I am not impressed.
Our EMC Symetrix fell over a couple of weeks ago due to a duff controller board and a situtation theat stopped fail over to the second controller, however no data was lost and in less than a day it was fixed.
Our Netapps 3020c has lost drives but rebuilds successfully each time so far as do all our HP servers (DL360's and 380's).

I've had a HPT372 based card for about seven years running a RAID 1 array. The only recoveries I've had to make were due to very odd power glitches or rudely behaved/badly documented Linux distro installers. In all cases just one of the drives was affected and could be rebuilt. Also fortunately, I could boot and make a backup of the 'good' drive before using the BIOS based rebuild utility; lesson #1 Never Boot from the RAID array!
For these fake raid chipsets, verify that the drives are individually readable when the array is broken; no problem with HPT37x. The HPT card has built in DOS INT13 support so the array can be tested and manipulated by standalone boots of Partition Magic not to mention gparted, testdisk, etc.

I also have a Promise Fastrak 100TX2 (2 ch PATA like above HPT) but haven't used it for RAID.
I'll be wary of the Sil3112 SATA RAID chipset on one of my m/b's due to the above thumbs down on the Sil3124.
Any reports on the latter two chipsets are appreciated.