Slashdot Mirror
Basics of RAID
Doggie Fizzle writes "RAID has been common in business environments for ages, and is now becoming more viable and popular for personal computers. This article focuses on the the basics of RAID, and spells things out for beginners or tech veterans. From the article: 'The benefits of RAID over a single drive system far outweigh the extra consideration required during installation. Losing data once due to hard drive failure may be all that is required to convince anyone that RAID is right for them, but why wait until that happens.'"
There's an excellent guide to RAID levels (with pretty diagrams and such) at http://www.acnc.com/raid.html
http://en.wikipedia.org/wiki/Redundant_array_of_in dependent_disks
RAID0 will increase your change of failure since you will loose all your data if a single drive fails. RAID0 isn't really redundant.
Cuz the boss won't cough up the money until it happens.
Slightly OT, but this site is frequently updated with the current state of SATA/RAID support under Linux.
I've had my data corrupted by a dodgy controller... twice. I've been having terrible luck with the nvRAID provided on nVidia nForce 4 motherboards. Twice now the machine has locked up and on rebooting the RAID array is reported as damaged and a drive is missing from the array. A bit of Googling has revealed it's a common problem.
Fiddling around in the BIOS disabling and reenabling RAID makes both disks show up again but putting them back into a RAID array seems to do no good as Windows always claims files are missing after doing this. If I reinstall Windows my data is always all still there in perfect condition, the hassle of reinstalling Windows and my apps is a pain though. So it's not totally corrupted, but enough to be a complete bitch.
My feeling on RAID on the desktop is that it's a good idea but at least in nVidia's case it's being done on the cheap and is not totally stable. That said Intel's RAID controllers are superb and I'd use one anyday if it weren't for the vast amounts of heat and inferior performance of the P4.
... but how often do personal backups actually happen? I'm one of those guys that has been taking home backups seriously for a long time, and has a collection of obsolete tape units to prove it. And backups still do not happen often enough if it requires me handling tape.
Let's face it, discipline is a drag, that is why at work IT people are paid to schlepp around stacks of locked cases full of back up tapes to be shipped off site.
So... for my home file server, I went to RAID mirroring, with a 3rd drive in a drawer. A mount-copy-umount chron job copies to the drawer-drive. Drawer-drive gets swapped and taken off site "when I think of it". Because... RAID only protects you from falling over hard drives. It does not proctect you from:
1) Ooops, I wish I hadn't deleted that.
2) Gack! My house just burned down! And took 10 years of tax data with it!
3) Power supply goes wonky, causing both drives to scribble random scorfulentness everywhere.
A home RAID system does not need to be expensive. Who needs hot swap? Use cheapo PATA drives. A few hours of down time for the wife and kids is OK. It doesn't take a big, bad CPU, and software RAID works great.
I've been using this board from Asus for about 6 months with onboard SATA RAID5. It cost $120 from Newegg.com when I got it, if my memory serves me.
I should probably add that SATA is closer in terms of functionality to SCSI than PATA. This does NOT mean that all controllers support it, however.
Looks ok to me - not a single one. Want my AdBlock list?
When you're setting up a RAID set using both striping and mirroring, do you want to set up two stripes and then mirror between the stripes (0+1), or do you want to set up mirrored pairs and then stripe those mirrored pairs (1+0)?
This is a quiz, and your data will grade you.
What you want, by far, is RAID 10 (1+0).
When you set up two stripes and then mirror across them, if you lose two disks, any disk in the first stripe and any disk in the second stripe, you lose all the data.
If you stripe across mirrored pairs, then the only way to lose data is to lose both drives in one of the mirrored pairs. You can lose any other disk than the second drive in a pair, or even many more disks, as long as they aren't both in the same mirrored pairs.
This doesn't make a difference with 4 drives. At 6 drives and up, use 10. Your data and users will thank you for it.
I am running RAID 5 on my desktop server right here. It has a P4 3 year old Gigabyte motherboard. It's not hotswappable because it's not enterprise level (and I don't plan on having to hotswap all of the time, only when shit happens) but it gives me the RAID 5 that I like to use as a backup using software based RAID on Ubuntu Linux. After the install, it it would be just as easy for Grandma to use as if it were not RAIDed and I am certain any /.er could figure out the install for most any Linux distro.
Can I have your money now?
It's amazing how common RAID is now, especially (S)ATA RAID.
In video editing, RAID is everything. External SATA RAID is the big thing now, and it works pretty well, even when it's OS based. What I haven't seen yet are (relatively) cheap SATA RAID 5 enclosures. That would be the Holy Grail of fast media storage.
Adblock must be doing it's job.
The man who trades freedom for security does not deserve nor will he ever receive either. - Benjamin Franklin
Ads...I see them once in a while, but they are quickly added to my adblock list. Really, should tech saavy /. readers see very many ads at all? I think the poster of the parent is a wanna be geek. Ok, enough sluphing off, back to work...
Yes and No.
For a basic RAID, ie a host card in a computer attached to a bunch of disks, then yes, the card is the single point of failure. This is less of a problem than losing a disk because: the card is less likely to fail (as you guessed, no moving parts) and failure of the card doesn't necessarily mean loss of any data. Failure of the RAID card will mean access to the attached RAID array becomes difficult =) however some machines even have hot-swap adapter cards, in which case you swap out the card for a new one and you're back up and running. No downtime, no loss of data.
For the usual kinds of RAID units used in enterprise, they're a hardware RAID enclosure with it's own RAID controller, attached via SCSI or FibreChannel to a host. Some of these units actually have two (or more) RAID controllers in them, each with their own FibreChannel connections (two or more, per controller) and if one controller dies, the other one takes up the slack (with a loss of performance, but not data). Going even further, you can have separate FiberChannel connections from each controller going through separate FC Switches, so even if a FC Switch dies, the SAN doesn't go down...
What's the downside to all this? Big $$$.
Cheers,
Kai
Specialist Mac support for creative pros, Melbourne
I worked for years in development of RAID solutions for a major manufacturer. One of the problems with selling RAID solutions is the lack of understanding, or the prejudice and bias of the people who were supposed to be specifying and buying the hardware.
The 'tutorial' of the parent article is talking in kindergarden terms, oversimplifications and obsolete term, and overlooking some of the issues with using RAID. It's a good example of the true lack of understanding about the subject. By now, there are so many types of solutions that the term RAID hardly applies. But, even 10 years ago companies like Compaq had innovative rudundant storage solutions that were enterprise ready.
Best regards.
1. RAID 0 is stripping. You are talking about RAID 1 and when your controller fails it doesn't take your OS and all your data with it.
2. RAID 5 runs fine on 3 drives.
Speaking of SCSI features, NCQ is now available on some SATA drives. It's not the full blown SCSI version, from what I gather, but it does bench well. See Tom's article on it.
I have two RAID controllers populated with two drives each in RAID-0. One has two of Western Digital's Raptors (74GB 10,000 RPM). The other with two of Maxtor's DiamondMax HDDs(250GB 7200 RPM). The latter has NCQ and benches significantly better. Some of the difference may be related to one RAID controller being better than the other (I didn't benchmark both controllers with the same drives), but I suspect the bigger difference is due to the NCQ features of the DiamonMax HDD.
I want this account deleted.
These days I think software raid is really the way to go, at least in comparison to the raid built into consumer grade raid cards. With software raid you should be able to move your disks to a working computer and boot up with a Knoppix CD and access your data if you have to. You can also raid individual partitions rather than entire disks. You could make a small non-raid boot partition on each disk which you sync regularly plus a larger data partition which becomes part of the raid.
The above applies to Linux, I don't think the non-server editions of Windows can do anything but raid 0 (maybe raid1?). Possibly a BartsPE CD could be used to recover a failed Win raid array.
You do know that nvRAID is fake, right?
As in, there's no RAID controller, it's software RAID done with BIOS code so that you don't have to dick with Windows as much.
Don't thank God, thank a doctor!
One glaring error:
RAID can be run on any modern operating system provided that the appropriate drivers are available from the RAID controller's manufacturer. A computer with the operating system and all of the software already installed on one drive can be easily be cloned to another single drive by using software like Norton Ghost. But it is not as easy when going to RAID, as a user who wants to have their existing system with a single bootable hard drive upgraded to RAID must start from the beginning. This implies that the operating system and all software needs to be re-installed from scratch, and all key data must be backed up to be restored on the new RAID array.
Again, wrong, wrong, wrong. There are hardware RAID 1 controllers that require no drivers and you don't have to do squat - just power down the server, install the RAID 1 on your IDE interface, plug in the new drive, hit the power, and away you go. The controller is smart enough to automatically sync up the two drives in the background.
-- Ed Carp, N7EKG erc@pobox.com PGP KeyID: 0x0BD32C9B What I'm up to: http://intuitives.mine.nu
I've been using software-RAID with ATA drives on Linux for quite some time, so I can comment on the behaviour of an array containing a faulty drive.
/dev/hdc, ATA error count increased from 0 to 1" and that it would be a good idea to check up on the hosts syslog.
/dev/md0 -a /dev/hdc1" the new one into the array, starting to resync it right away.
First off, let me emphasize how important it is to set up proper email notification (or pager etc.) for such cases! If you don't know about the failure, you're certain to get nice phonecalls from affected users.
If you've set up the notification system (smartd and mdadm come to mind), you'll eventually get an email saying something like "Device:
Checking up on the system, you'll find that the average system load has increased substantially, which is due to the system trying to persuade the disk to write to a faulty sector and the software RAID having to compensate, queuing the errors.
Depending on how often the defective sector is tried to be written to, the load can increase to values of 10 and above, rendering the system unusable. This is a good time to halt it and replace the defective drive, partition it and "mdadm
This may sound really horrible, but in practise it's usually less 60 minutes (counting from receipt of the first notification email) until normal operation can resume with such a system. This is assuming you have all spare parts stored somewhere on site.
In genereal, I've found software RAID1 and software RAID5 on Linux to be exceptionally stable. I'm also very happy about the performance, given that all I'm using is a bunch of el-cheapo ATA disks. As for reliability, I'm convinced it can't be beat in the consumers' price range, since I've seen too many consumer grade hardware RAID controllers go down in a swirl when putting more than a light load on them.
In the enterprise, I've seen companies move to software RAID on their Linux systems, because they found out that their only 5 years old enterprise hardware won't be getting any new spare parts anymore, which includes motherboards, CPU and IO controllers. Moving to software RAID on enterprise grade SCSI stuff allows them to move the entire system to another piece of hardware simply by moving the harddisk to it.
Forget all that. This is an excellent article on RAID.
http://arstechnica.com/paedia/r/raid-1.html
Consider:
RAID 10 disadvantage: "All drives must move in parallel to proper track lowering sustained performance". In fact each drive can seek independently for reads and only pairs must seek together for writes.
RAID 1 advantage: "Transfer rate per block is equal to that of a single disk"
RAID 5 disadvantage: "Individual block data transfer rate same as single disk"
Would be nice if it was consistent about whether that's good or bad.
RAID5: "Highest Read data transaction rate" except for RAID 10, of course, where you've less chance of being bottlenecked because there are two sources for each stripe.
RAID5: "Medium Write data transaction rate", only the lowest of all except 50, because of the parity calculating and writing to a second drive.