RAID Trust Issues — Windows Or a Cheap Controller?

NicApicella writes "My new system has two sparklin' SATA drives which I would like to mirror. After having been burned by a not-so-cheap, dedicated RAID controller, I have been pointed to software RAID solutions. I now stand in front of two choices for setting up my RAID: a Windows 7 RC software RAID or a hardware RAID done by the cheap integrated RAID controller of my motherboard. Based on past experiences, I have decided that only my data is worth saving — that's why the RAID should mirror two disks (FAT32) that are not the boot disk (i.e. do not contain an OS or any fancy stuff). Of course, such a setup should secure my data; should a drive crash, I want the system up and running in no time. Even more importantly, I want any drive and its data to be as safe and portable as possible (that's the reason for choosing FAT32), even if the OS or the controller screw up big time. So, which should I choose? Who should I trust more, Microsoft's Windows 7 or possibly the cheapest RAID controller on the market? Are there other cheap solutions?"

Comments

Re:Real RAID is cheap

[Grishnakh]

Wrong. You need to buy at least two of these controllers, at the same time, or else when your "real" RAID card dies (and they do), you'll lose all your data unless you can find an identical card (you may even need the exact same firmware version).

Software RAID on Linux is a much better solution, as the underlying hardware doesn't matter. You can mix and match different drive models/sizes (can't do that on HW RAID), and swap the drives to a different system and still read them thanks to the standardized on-disk data format.

Re:You are asking the wrong question.

[stfvon007]

Ive had #4 happen to me. A power supply in my computer failed (a name brand one, not a cheap no name brand) and damaged everything attached to one of the 12v rails. This included both drives of a raid1 set. (ironically all my drives that wernt part of a raid set were completely undamaged) I was later able to recover the data from both drives (both had damaged sections but different areas were damaged on the 2 drives allowing for a complete recovery between the 2 of them)however it goes to show that just having a raid array wont completely protect you from hardware failures.

Cheep Non-RAID Controller!

[Tux2000]

Fast facts:

• NTFS and ext3 have journaling, FAT12/16/32 and ext2 don't have journaling.
• FAT12/16/32 have a central structure (the FAT). Damage it and your data is lost. ext2 and ext3 store their meta data redundantly.
• RAID is no replacement for Backup.
• A real hardware RAID is expensive, and appears to be a single disk to both BIOS and OS. Its on-disk meta data is propritary, i.e. if your HW RAID controller dies, you need exactly the same controller again to get access to your data. HW RAID works with every OS, because it appears to be a single disk (typically, SCSI). Booting from complex RAID configurations is no problem, as each RAID appears to be a single disk. The RAID controller is a small computer on its own, taking care of the reqired calculations for non-trivial RAID levels, of switching to hot-standby disks, and of detecting broken disks.
• A software RAID is cheep as dirt, every single disk of the RAID appears in BIOS and lower levels of the OS. The on-disk meta data depends only on the OS, so you can mix controllers as you like. A broken controller is no problem, replace it with any controller that has the same connectors and your data is back. Booting can be a problem, because the BIOS does not know anything about the RAID. Usually, booting is only possible for RAID-0 and RAID-1. Booting another OS is problematic, because there is no standard for Software RAIDs. Linux may be able to work with Windows RAID volumes, but Windows can't work with Linux RAID volumes. Calculation and monitoring is done by the host CPU.
• A host RAID is nearly as cheep, the only difference to a software RAID is that the BIOS decides about the on-disk meta data. Special drivers for each supported OS know the structure of the meta-data, but they don't allow to use other controllers in the same RAID. A broken controller is a problem, because drivers will refuse to work with other controllers. Booting is no problem, because the BIOS knows about the RAID.

I prefer pure software RAIDs, for a simple reason: They do not depend on available hardware. If one controller dies, switch to another one: Other brand, other type, other drivers, and the RAID still works. If you insist, you can even mix an IDE drive, a USB drive, a SATA drive and a SCSI drive into a single RAID. Try that with a hardware or host RAID. Some people even built RAIDs of floppy disks or USB sticks (not for pemanent use, of course).

My faithful old Linux home server runs two RAIDs, both in software: a RAID-1 for the OS (remember: the BIOS does not know about the RAID), and a RAID-5 for the data. The RAID-1 used to run on old SCA drives, but recently, I switched to two small IDE drives due to unrecoverable SCA cabling problems. The RAID-5 is composed of four IDE drives, connected to two IDE controllers, each disk on a single IDE cable. An external USB disk is used to back up my data, rotating through 10 days. All filesystems are ext3, all disks are monitored using SMART, all RAIDs are monitored. If anything wents wrong, I will get an e-mail from the monitoring software.

Until recently, one of the controllers was an el-cheapo non-RAID controller, and the other one was a donated, expensive, well-known brand, RAID-capable controller running in non-RAID mode. The latter one decided to randomly take some free time on the job, and either disconnected from the PCI bus or disturbed it, causing panics in the OS above. Only pure luck protected me from data loss. I ripped it out of the machine, kicked it into the trash bin, rewired the RAID to use two disks per IDE cable, and verified and reconstructed my data. Some days later, another el-cheapo non-RAID IDE controller arrived, the same brand, model and type that already sat in the next PCI slot. So I rewired the RAID again to work with one disk per cable, everything was fine again.

For a new small business or home server, I would use nearly the same setup again: Two software RAIDs, one for the OS, and one for the data. Upgrading the OS is just fun when you can

Re:You are asking the wrong question.

[Ant P.]

That's an interesting thing... the power supply has more potential to cause damage than anything else in the PC, but nobody ever thinks about protecting against its failure. Makes me wonder why we don't have surge protectors on the 5/12V rails as standard yet.

Re:Are you crazy?

[hedwards]

I'm not sure why the parent and GP were modded funny, I refuse to buy drives by Maxtor or WD these days because of the crap quality of the past. Sure it's been long enough that they probably have fixed the issue, I just don't trust them, even running mirrored ZFS.

I'm still somewhat astonished that WD would think that it's acceptable to have external drives that work on OSes other than Win except for the power management features. Saying you're just supporting Win for a hard disk is nowhere near acceptable.

Personally, what I do at home is I use ZFS to mirror a pair of 1tb Seagate drives and that seems to work fine, it's not really the best set up, but it's hard to get such things located off site for the amount of money I have to spend.

Re:You are asking the wrong question.

[Theovon]

Heh. Linux software RAID doesn't do jack. I've looked at the source code. The mdadm RAID1 driver just alternates drives for reads whenever the requests are not contiguous. That is all. Nothing more. There's no intelligence in there. No keeping track of head positions, no attempts to discover or infer physical drive geometry. Nothing. Just a simple round-robbin. It just so happens that for MOST things that involve random access, the effective throughput is nearly doubled. More intelligence wouldn't actually buy you much in the general case, so why bother?

Also, the dmraid (fakeraid) RAID1 driver only does reads from one disk. I made the mistake of using dmraid instead of mdraid, only to discover through performance tests and iostat that there are basically two software RAID drivers that CLAIM to do identical things but in fact do not.