RAID Hard drives

[Dave]

I really don't get this raid business...

Is RAID where you can have two drives that save exactly the same information on them.. so basically if you save to one, you save to a back up?

So what is it called when you combine more than one hard drive so that they act as one?

[Mook]

There's a few different types of RAID, all of which allow you to use more than one hard disk.

RAID 0 - normally uses two disks and the data is written across both but it's not mirrored

RAID 1 - where there is a mirrored pair of disks and this provides redundancy, in case on of the disks fails (it's also a lot quicker to write data to compared to RAID 5

RAID 5 - needs three or more disks and allows you to build a bigger "virtual disk" for want of a better term. So if you start with 3 x 500GB disks and make a RAID 5 array, you would have a 1000GB logical drive, because one of the disks is used for redundancy. It's very scalable, though - the limit on the number of disks is restricted by the RAID hardware. If you wanted a 2000GB logical drive, you could do this with 5 x 500GB hard disks - 4 x 500 = 2000, plus one of the disks would be used for redundancy.

[Dave]

Thanks mate.. but why would you want a redundant disk?

take your raid 5 example.. you go out and buy 3x500GB drives and in effect can only use two of them

[shark_90]

It's important to say that the space equivalent to one whole disk is used for redundancy in RAID 5, not an individual disk. An individual disk is RAID 3 which is utter rubbish

[shark_90]

[ QUOTE ]

Thanks mate.. but why would you want a redundant disk?

take your raid 5 example.. you go out and buy 3x500GB drives and in effect can only use two of them

[/ QUOTE ]

That's right, but one of them can fail and your system still works. It's expensive but it's redundant. If you lose a drive that's not in a RAID 5 array, and you lose the data on it, how much would it cost you in downtime if you were a business?

Usually worth the investment

[Mook]

[ QUOTE ]

take your raid 5 example.. you go out and buy 3x500GB drives and in effect can only use two of them

[/ QUOTE ]

You would use this for redundancy, in case one of them fails. Unless you're a reall propellor-head, you wouldn't have this set-up on your home PC. RAID5 is normally used in a server environment, where the server has to be up and running all the time. If you have one disk in a server and it fails, your server is dead. If you have a RAID1 or RAID5 array, if one of the disks fails, the server keeps running.

To give you an example, the servers we now run in our datacentre have both RAID1 and RAID5 - the operating system runs on the C drive using RAID1 with two drives (we don't need much space for this, but we need redundancy) and the storage (all files etc.) is on the D drive using RAID5 with 8 drives (we need shed loads of space for some of our customer data).

Is this making sense?

[Dave]

Ok, still don't quite understand.

So in raid 5 you have 3 disks, one of which is redundant.. if one of the disks goes down you still have all your information?

So how does that work?.. you have the data saved twice on different drives?

[Mook]

Yes, you still have all of your information

The data is "striped" across the drives, so, for each file, there is part of that file on several logical drives. The additional drive knows where all the bits are.

[shark_90]

It's better to draw a picture but I'll try to explain.

RAID 5 works by striping data across all the disks in your RAID 5 array.

Say you have 3 disks. It writes a stripe of data to disk 1, then a stripe of data to disk 2, then a stripe of parity to disc 3.

The next operation would be data to disk 1, parity to disk 2, data to disk 3.

The next operation would be parity to 1, data to 2, and data to 3.

So you get:

D - D - P

D - P - D

P - D - D

D - P - D

D - D - P

And so on. Basically if you lose a drive the parity stripes can be used to work out the data that should be on the lost drive, and allow the system to keep running until such a time you can exchange the dead disk for a new one, where it would then proceed to rebuild the disk automatically.

[Dave]

Right! getting there..

Ok, you have three drives and bytes are written to each. i get that..

But say drive 1 goes tits up.. completely seized.. how can the other two drives compensate for the information lost on that drive?

[shark_90]

Because of the parity stripes across the two remaining working drives. The RAID controller can unravel the parity bits and work out the data that was on the dead drive. It reduces performance while it is doing that, but it means you can keep working in the short term.

Clever eh?

[Mook]

Just Googled "RAID explained" and it spat out this:

http://www.ahinc.com/raid.htm

[Dave]

[ QUOTE ]

Clever eh?

[/ QUOTE ]

Very!!..

and even if that data is data rich, say as in a photograph, it can re-produce it exactly as it was?

That is very clever

in that case why is it not the industry standard?.. does it come down to expense?

[Dave]

[ QUOTE ]

Just Googled "RAID explained" and it spat out this:

http://www.ahinc.com/raid.htm

[/ QUOTE ]

Very good piece!!.. thanks

[Dave]

Ok,

So, do you have to install a RAID set of hard drives at the set up of a new computer, or can you add a new drive and make it a mirror of the old drive?

[shark_90]

Yep, it's completely irrelevant what data it is, as it's written as raw blocks anyway.

I guess it is the industry standard for storage volumes, but it depends what is required. RAID 0 is much faster (the fastest) so if you're moving a lot of data quickly but don't care too much about redundancy then RAID 0 is the way to go. If you want speed and redundancy you can use RAID 0+1 (also called RAID 10) but that's even more expensive as you lose 50% of your useable disk space instead of N-1.

[shark_90]

[ QUOTE ]

Ok,

So, do you have to install a RAID set of hard drives at the set up of a new computer, or can you add a new drive and make it a mirror of the old drive?

[/ QUOTE ]

It depends. You can use software RAID (in Windows for example) where you can use Dynamic Disks which allow you to create a mirror of a drive at any point. Software RAID is generally slow though.

Or you can use Hardware RAID (i.e. have a RAID controller in your computer that the disks are plugged into). Hardware RAID is faster and more reliable - but I can't say I've tried to add a disc into an array and mirror it with data on already. There may be some spangly RAID controllers that will do it, but typically you can't.

However, with a SAN (that's loads of disks in a huge 19" enclosure) you can do just about whatever the hell you like

[Dave]

Thanks guys..

so is this how sites like e-bay and amazon keep their up time so good

[shark_90]

They would without doubt be using some huge SAN environment, but basically yes

Ask Bison what he uses for TSN. Infact, why don't you convince him that it would be good if you went to see it

[Mook]

Well, yes, that's part of it. They also run network load balanced (NLB) web servers and clustered database servers, all of which will doubtless use RAID arrays on SANs, but NLB and clusters are the topic of a whole different thread!

[Cupramax]

[ QUOTE ]

Thanks guys..

so is this how sites like e-bay and amazon keep their up time so good

[/ QUOTE ]

Any business critical stuff should be either Raid5 or mirrored as an absolute minimum, end of... Big web sites would more likely have a clustered solution i.e. multiple servers taking the loading but these would no doubt use mirrored disks as well.

[CarMad]

For people like Amazon / play etc. its not the data thats the problem but the performance / resilience of the server. They don't normally have huge amounts of data on a SAN as such but they will often have dual data center set ups and multiple server streams that get load balanced.

PLAY.com for instance has about 20 seperate servers that run on the back of it. Some are to help database searches that come up with the products, the rest do various things like server out the pages or process the encryption etc.

[Rachel]

[ QUOTE ]

Any business critical stuff should be either Raid5 or mirrored as an absolute minimum, end of...

[/ QUOTE ]

"end of..." indeed. I suggest that for a Friday brain mulcher you go have a read of this:

http://www.intel.com/design/storage/raid6.htm

[Rachel]

[ QUOTE ]

For people like Amazon / play etc. its not the data thats the problem but the performance / resilience of the server. They don't normally have huge amounts of data on a SAN as such but they will often have dual data center set ups and multiple server streams that get load balanced.

[/ QUOTE ]

Hopefully I'm about to build a non-stop dual site business continuity environment in the next few months.

The trick with SANs is the "fabric" and whether you use standard litle switches or go insane and use intelligent director class devices to run the fabric (or storage network), and then whether or not you go the whole hog, enmesh your "SAN islands" and virtualise your storage.

You can find a bit more light reading here or go to say the Cisco website and look up the 9500 series MDS, or the latest from EMC.

[shark_90]

It's not exactly mainstream or supported by many vendors looking at that list.

I don't see the advantage of that over RAID5 + Hot Spare

Plus you should have a decent support contract in place anywhere that RAID 6 would be viable, so a failed RAID 5 drive should be replaced straight away anyway.

Marketing bobbins in my opinion, I just can't see it's use