How does RAID work?

aplefka

Lifer
Feb 29, 2004
12,014
2
0
I've read up a lot on it but I still don't understand it. How does RAID make it possible to access data faster? What's the difference of running a hard drive in a RAID setup than if it were just stand-alone?
 

hippotautamus

Senior member
Apr 10, 2005
292
0
0
RAID is all about protecting large amounts of important data. You put multiple drives on an array so that if one fails, you have a backup all the time and you can rebuild the data quickly.
 

Crusty

Lifer
Sep 30, 2001
12,684
2
81
It all depends on the type of raid.

RAID 0 has no redundancy at all, one drive in the array fails and you lose it all.
RAID 1 is a mirror. If one drive dies then you can still operate until you replace the array.
RAID 5 is a more efficient RAID then 1 in terms of storage space, but you sacrifice speed. It also uses parity instead of just mirroring.

http://www.anandtech.com/guides/viewfaq.aspx?i=110

That has a LOT of information about RAID, and the different kinds as well as the tradeoffs of using the different levels of raid.

I personally prefer RAID 5 over the others. I use it to protect data, not for speed.
 

aka1nas

Diamond Member
Aug 30, 2001
4,335
1
0
Raid 0 is the only one that gives you extra speed without losing capacity. However, you have no data redundancy with this mode. It is still nice for non-critical data, for example if you use an array for capturing video and can use the extra speed.
 

Fullmetal Chocobo

Moderator<br>Distributed Computing
Moderator
May 13, 2003
13,704
7
81
I've often seen RAID 5 configurations drastically improve read performance. Moreso with RAID 50 and 50n arrays as well.
Tas.
 

Calin

Diamond Member
Apr 9, 2001
3,112
0
0
Even a RAID1 (mirroring) can improve read speed - you can read some data from one drive and some data from the other drive. However, some implementations will simply read data from both disks and compare it to see that it is the same.
 

Steve

Lifer
May 2, 2004
15,946
11
81
When you stripe data, say across two disks, you read 50% of the file from each disk rather than all of it from one.
 

redhatlinux

Senior member
Oct 6, 2001
493
0
0
It's A Redundant Array of In-Expensive Disks.. RAID has several varieties, GOOGLE is your best Friend ... try it.
 

Smilin

Diamond Member
Mar 4, 2002
7,357
0
0
Originally posted by: booradley
Originally posted by: redhatlinux
It's A Redundant Array of In-Expensive Disks..

Ive heard that said a few times now and never got the joke... Someone wanna enlighten me?

It's not a joke. That's what the acronym stands for. RAID was originally designed to make a fast and reliable disk out of leftover junk drives.
 

Smilin

Diamond Member
Mar 4, 2002
7,357
0
0
Originally posted by: aplefka
So then RAID is more for backup instead of getting faster speeds out of drives?

No, RAID is never for backup. It's for redundancy. Raid is designed to protect your uptime, not your data. Backups are for protecting your data and there is no substitute. Raid simply means you won't have to down the server when you lose a drive.

The speed is a side effect of how the system is implemented.
 

MrDudeMan

Lifer
Jan 15, 2001
15,069
94
91
backups are mostly put on tapes. tape drives are very good for backups since you can fit a ton of data on one. they are slow, but mind you this is only for backup, not for constant read/write. RAID is to be used for up-time, like Smilin said, not for data security. well, it is data security, but not very good security. having several backups for important information is the best way to make it secure. ive always gone by the rule that nothing is saved for real unless it is in at least 4 different places on at least 2 different kinds of "permanent" media (i.e. a dvd and a tape, or an external hard drive and a tape and a dvd, or whatever combination you want).
 

Steve

Lifer
May 2, 2004
15,946
11
81
Also keep in mind that there is a difference between backups and archiving. For the latter you want something optical, but not just CD or DVD. Try looking at MO drives (magneto-optical).


Back on topic, RAID is all about uptime, and performance benefit is a bonus, as Smilin said.
 

TGS

Golden Member
May 3, 2005
1,849
0
0
Getting slightly off topic from raid, but for archiving you rarely will want to use any magnetic media for long term archiving. The electric charge of the tape media will dissipate over time and leave you with garbage data sets.

RAID is indeed meant for disk redundancy, though the way it can make data transfer faster is the fact that the array of multiple disk with utilize more drive actuators. This lessens the mechanical latency of the individual drives. For most RAID types. Some RAID types give better data protection schemes, better read performance, better write performance, and some a mixture of all.
 

iamspartacus

Junior Member
May 27, 2005
5
0
0
RAID: random array of inexpensive disks

RAID can be used either to back up the same data on multiple hard drives. So you are writing one peice of data 4 or 5 different times depending on the setup. Or you can have one peice of data split up and written in the same sectors on the hard drive. So one file is in the same sector on multiple disks allowing for faster reading.
 

imported_Phil

Diamond Member
Feb 10, 2001
9,837
0
0
Originally posted by: sm8000
When you stripe data, say across two disks, you read 50% of the file from each disk rather than all of it from one.

If, and only if, the controller supports it. Most consumer-level controllers don't, but a decent, intelligent controller will.

For instance, the two-drive Promise PATA RAID card (I forget the model) reads from one disk only. I've sat and watched it :)
 

imported_Phil

Diamond Member
Feb 10, 2001
9,837
0
0
Originally posted by: Smilin
Originally posted by: booradley
Originally posted by: redhatlinux
It's A Redundant Array of In-Expensive Disks..

Ive heard that said a few times now and never got the joke... Someone wanna enlighten me?

It's not a joke. That's what the acronym stands for. RAID was originally designed to make a fast and reliable disk out of leftover junk drives.

Wasn't it changed a while ago to mean "Independant Disks", as the disks themselves aren't actually all that inexpensive?
 

TGS

Golden Member
May 3, 2005
1,849
0
0
It's call concurrent reads, IIRC.

Originally the term was coined, for inexpensive arrays. People typically use the term in independant disks do to the fact that hard drive arrays are no longer ridiculously expensive to build.

Link of Origins, etc...
 

Valkerie

Banned
May 28, 2005
1,148
0
0
Hard drives are hella slow and they are one of the most extreme sources of bottlenecks in a comp.

So with

RAID 0 - performance that kills tonz of bottlenecks, 'specially for playin' games!

RAID 1 - mission critical, protects data from HD failures

RAID 5 - faster and far more secure than 0,1

Whatever, these are basics.

P.S. Your rigs will own any games at more anticipating levels with RAID 0, because it then comes down to CPU, memory and video card - performance, rather than worry about HD bottlenecks.
 

aiex

Senior member
Jul 5, 2001
914
0
0
Just to add my own 5c, RAID 0 isn't actually RAID at all since there is no redundancy.

Ai3x :D
 

TGS

Golden Member
May 3, 2005
1,849
0
0
Originally posted by: Valkerie


RAID 0 - performance that kills tonz of bottlenecks, 'specially for playin' games!

...


RAID 5 - faster and far more secure than 0,1

P.S. Your rigs will own any games at more anticipating levels with RAID 0, because it then comes down to CPU, memory and video card - performance, rather than worry about HD bottlenecks.

RAID 0, really doesn't kill any bottlenecks in gaming other than load times. A RAID 1 setup with concurrent reads should accomplish the same task, with real data redundancy, though with half the available space.

As far as I know, I cannot think of a single integrated RAID controller that supports concurrent reads. At least of recent AMD chipsets.

RAID 5, is not faster than RAID 1, in the fact that 5 must calculate parity bits per write operation. RAID 1, 0+1, and 1+0 does not have to make any type of calculation and data is striped or mirrored to the appropriate physical devices/locations.

For cost per usable storage volume, RAID 5 is by far the best option. Protection, High Volume usage, slower write performance. For large scale cost is no option arrays, 1+0 is for serious I/O environments were data CANNOT be lost. Beyond a 1 or 1+0 setup you are looking at remote mirroring which is beyond the scope of RAID.

Also another factor in RAID 5 vs 1, 0+1, 1+0 is rebuild time. RAID 5 takes considerably longer to rebuild do to having to recreate data through the parity bits. RAID 1, etc... does data set copies which means faster rebuild performance. After all you are using RAID so you can rebuild and continue to access your data. Why would you want a significant performance hit when one device goes out? :)

Just an FYI as well, 0+1 is not the same as 1+0. 0+1 can sustain the loss of 1 drive per stripe before the array goes Tango Uniform, while 1+0 can lose the mirrored set of each device before it goes down.

linky
 

aka1nas

Diamond Member
Aug 30, 2001
4,335
1
0
Originally posted by: Valkerie
Hard drives are hella slow and they are one of the most extreme sources of bottlenecks in a comp.

So with

RAID 0 - performance that kills tonz of bottlenecks, 'specially for playin' games!

RAID 1 - mission critical, protects data from HD failures

RAID 5 - faster and far more secure than 0,1

Whatever, these are basics.

P.S. Your rigs will own any games at more anticipating levels with RAID 0, because it then comes down to CPU, memory and video card - performance, rather than worry about HD bottlenecks.


Wow. Umm, not quite. RAID 0 does very little to help you "0wn" in games other then in loading levels a little bit faster. This is because no form of RAID improves the latency(seek time) of the hard disk drive. RAID can increase transfer rates, which is only useful when reading large amounts of data in a short period of time(i.e. loading levels). Most of the time what makes the hard drive bottleneck is that it takes milliseconds to access it instead of the nanoseconds it takes to access data in RAM.

RAID 1 offers basic redundancy with no performance penalty but with only 50% storage efficiency.

RAID 5 uses 3 plus drives and distributes parity data across the entire array. The point of this is that the array can lost a certain amount of drives without losing data. This is better than standard RAID 1 as you still get some extra read performance due to being able to read data concurrently from multiple drives.
With RAID 1 you can potentially still lose the entire array if 2 drives(the same drive on each side of the mirror) were to fail together.