RAID 0 with more than 2 drives

[joaoparaiso]

Hi,

My computer has two Hitachi 7K160 (the 1-platter sibbling of the 7K500) SATA-II drives in RAID 0 configuration. It provides top-notch performance (in the 7200rpm realm, only the new 7K1000 in RAID 0 beats it) and I'm very satisfied with it, but now I need more disk space.

I wonder if I can just buy another of these drives and make a 3 or 4 disk RAID 0 array. Is it possible? Will there be any performance gain (even if marginal)?

By the way, is it compulsory that all the hard drives in a RAID 0 array must be equal? I mean, could I use, for instance, a 250 Gb Hitachi hard drive and add it to the current array? Could I use, hipotethically, a drive of other brand (Seagate, etc.)?

 

[Smilin]

Raid 0 spec allows for as many disks as your controller can handle. Beyond 2-4 though and cheapo controllers won't cut it.

The performance is (very roughly) determined by the number of disks. Theoretically a 4 drive raid 0 array will be twice as fast as a 2 drive raid 0 array. It will also have an increase in suceptibility to failure. If you lose one drive you still lose the data on all 4.

You can use different drives.

Space available will be (number of drives) x (size of smallest drive).
Speed will be (number of drives) x (speed of slowest drive).

There is some overhead in all this so it's not 100% exact. Close enough for practical concerns though.

 

[NikolaeVarius]

I'm not sure about speed. In the Latest reviews, Raid 0 increased performance by like 2 seconds. In some cases it lowered it.
Theoretical numbers are not Real World numbers

Also, I would deem it risky to have multiple RAID Drives. If one fails, prepare to cry.

 

[Smilin]

Originally posted by: tenshodo13
I'm not sure about speed. In the Latest reviews, Raid 0 increased performance by like 2 seconds. In some cases it lowered it.

What does 2 seconds mean?

For sustained througput it will be very similar to the theoretical numbers.

For random I/O the seeks times don't change so you won't get the same increases as you do with sustained throughput. On servers where massive amounts of caching occurs sustained speed can rapidly fill the cache which then offsets random i/o to some degree (random makes for poor cache hit ratio. less cache worse it is). On desktop systems with cheap assed controllers the random i/o greatly reduces the benefit of raid. There is still a benefit though.


The added risk of raid is also negligible for 2 drive arrays. With 4 you might be slicing up your MTBF on the drives enough that you'll hit it before the drive becomes obsolete.

The long term safety of your data also has nothing to do with the reliability of your disks you know All disks should be expected to fail.

 

[NikolaeVarius]

http://www.anandtech.com/storage/showdoc.aspx?i=2969&p=8

2 seconds as in Load times for actual Applications.

 

[Smilin]

Originally posted by: tenshodo13
http://www.anandtech.com/storage/showdoc.aspx?i=2969&p=8

2 seconds as in Load times for actual Applications.

Ok so around 6% in those cases. That sounds reasonable. Lots of random i/o and lots more things happening besides i/o in general. You aren't just reading the levels off the disk you're also moving textures over to GPU memory such. There will be bottlenecks involved beyond the maximum throughput of the individual disks.

For sustained I/O though the numbers remain pretty close to the theoretical. 98% increased minimum, 41% increase in average...although that average increase is being dragged down a bit since it includes performance of a nearly empty disk which isn't common. That's on a pretty crappy controller though (typical of a home pc). You can tell by the burst rate and CPU usage. The controller is also capping the max which you can tell by the relatively flat graph:

http://www.anandtech.com/storage/showdoc.aspx?i=2969&p=4

I'm nit picking though. There is a definately a difference. Nothing huge though.

 

[Madwand1]

Large file access / file transfer is a real application, and even with some "free" RAID 0 implementations, it's possible to see great scalability with increasing number of drives. It's possible to screw up this very basic feature of RAID 0 -- STR -- and conversely it's possible to get this right with even the most basic implementation.

E.g. 2xRAID 0 vs. 3xRAID 0 performance measurements, with lowly old-generation 300 GB, which individually bench at around (65, 54, 35) MB/s (max, average, min).

http://i89.photobucket.com/albums/k203/Madwand0/nvr0-2drive-vs-3drive.png

Here's a chart showing how these numbers scale:

http://i89.photobucket.com/albums/k203/Madwand0/raid0-drive-scalability-1-3.png

The average numbers are most reliable, and they show near-perfect scalability. This is not a safe assumption, because sometimes slightly funny things can happen when you go from an even number of drives perfectly divides the access size (typically 64 KiB) with respect to stripe size to an odd number of drives, for which there is no stripe size which will perfectly divide the access size.

Some RAID implementations are bandwidth-limited in practice, which by definition limits the maximum STR possible. This is obvious and to be expected for standard PCI controllers, but there are also some surprises with recent nVIDIA motherboards / chipsets showing such behavior. My figures here are from an older nVIDIA nForce 3 board.

Unfortunately I don't have 4-drive figures recorded at present with this chipset and the same drives, but I recall another nForce (430) implementation which had STR > 250 MB/s, so the 200 MB/s or so apparent cap with the above measurements is not a hard cap.

 

[corkyg]

As you increase the number (n) of drives in the RAID 0 array, the stnstical probablility of a drive failure is multiplied by n. Any failure of a RAID 0 component is a total failure of the array.

Any gain in performance, IMHO, just isn't worth it.

 

[SoulAssassin]

Originally posted by: corkyg
As you increase the number (n) of drives in the RAID 0 array, the stnstical probablility of a drive failure is multiplied by n. Any failure of a RAID 0 component is a total failure of the array.

Any gain in performance, IMHO, just isn't worth it.

Agreed, if you have 2+ drives and have a hardon for doing RAID at least do RAID 5. You only lose n-1 in disk space and it will tolerate failure.

 

[Smilin]

Originally posted by: SoulAssassin

Originally posted by: corkyg
As you increase the number (n) of drives in the RAID 0 array, the stnstical probablility of a drive failure is multiplied by n. Any failure of a RAID 0 component is a total failure of the array.

Any gain in performance, IMHO, just isn't worth it.

Agreed, if you have 2+ drives and have a hardon for doing RAID at least do RAID 5. You only lose n-1 in disk space and it will tolerate failure.

not agreed.

The gain in performance is very much worth it.

First the statistical probability of a RAID 0 failure does increase but it is NOT (single drive failure rate) x (number of drives).

You must include the controller, psu, software used and system as a whole. One of the single points of failure on a Raid 0 array, the controller does not get eliminted if you are only using a single drive. If your system is hard rebooting from a software or hardware failure you run a risk of filesystem corruption on both a single drive and raid volume.

Also, RAID was spec'd in an era when drives did fail all the time. Hence Raid 1 & Raid 5. That was nearly two decades ago.

The current MTBF on a WD Raptor WD1500ADFD is 1.2 million hours. Here's how that math breaks down. 1,200,000 hours / 24 = 50,000 days = 136 years. Throw that in a four drive RAID 0 array (divide by 4) and it's a 34 year MTBF. Now this doesn't mean you should expect it to run for 34 years. That's not quite how MTBF works. A more accurate picture would be if you ran 34 drive arrays (136 drives total) for a year you could expect one of them to fail.

Just how long do you really want it to run? Statistically it's going to be full and/or obsolete before it fails. I'm living proof. I've got a RAID 0 array that has outlasted several other single drives that have also been in my system. There are now faster single drives on the market so it has been reliable beyond it's useful lifetime.

Besides, like I mentioned earlier. Running a single drive, RAID 0, or RAID 5 makes zero difference on how safe your data is. RAID 5 (and to a lesser degree a non 0 solution) is not designed to protect your data at all. It's designed to protect your uptime.

Backups protect your data. If you have them you can run RAID 0 to your heart's content and not have to worry a bit.