RAID on home desktop systems
About
Basically this a quick look at using RAID 0 and RAID 1 on desktop systems, as I only have enough hard drives to test these. There is no definitive guide which states that RAID must never be used, nor that a standalone hard disk is essentially "obsolete", especially given that so many motherboards now support RAID in some form or another.
The setup
The machine has an MSI P7N Diamond motherboard with a nVIDIA 780i northbridge and nVIDIA 570 southbridge. RAID support is:
- RAID 0
- RAID 1
- RAID 0+1
- RAID 5
- JBOD/Spanning
As stated above, only RAID 0 and RAID 1 are tested due to lack of hard disks. Additionally, the machine has an Intel Quad Core Q9300 @ 2.5GHz, 8GB of DDR2 1066 DRAM (set up as stock DDR2 800 in BIOS), 2x nVIDIA GeForce 8800GTs in SLI (not that this should make a blind bit of difference) and the adequate PCI-E SBlaster X-Fi Xtreme Audio card which comes with the motherboard (ditto).
Initial thoughts
Opinion is somewhat divided as to whether there is any real point in using any form of RAID setup on desktop systems. Some people believe it is a marketing ploy to sell more machines (i.e. bragging rights I suppose, if you're into that sort of thing), or that it simply isn't reliable enough to warrant the expenditure (on additional hard drives).
Irritatingly, some articles tend to contradict other, arguably more "informed" sources by rather wishy-washy conclusions that have succumbed either to marketing or possible bias.
Motherboard RAID options explained
RAID 0
This is where data is written to two or more hard drives in parallel. The file is roughly split so that part is written on one drive and the remainder on the other drives. Theoretically this should have fast read and write speeds as multiple hard drives are essentially performing different reads or writes on the same file simultaneously. This is borne out by "back of a fag packet" tests using HDTach (see below). RAID 0 is NOT strictly a RAID setup as there is no redundancy, i.e. if one hard drive fails, then you lost access to everything (short of paying a lot of money to professional data recovery experts). Also referred to as "Striped" arrays.
RAID 1
Two or more hard drives write exactly the same data in the same physical locations: this is also why it is referred to as "mirroring". Has benefits in that if one hard drive fails, you simply pop a replacement in the RAID setup, wait for the array to be rebuilt (which should be able to be done while the system is running) and you're back to normal. There should also be increased read speeds (as you have more than one hard drive which can search for the data) but this is offset by write speeds the same, or perhaps worse, than the standalone hard drive.
RAID 0+1
Requires at least four hard drives: two for RAID 0 and two more to mirror this array to the other two drives. Supposedly has all the read and write benefits of RAID 0 with the redundancy of RAID 1. This is true provided that only one drive fails. If two fail, you'd better hope it's the two drives that are part of one RAID 0 array...
RAID 5
Requires three or more hard drives: data is striped across all three but additionally extra information is written regarding the information for each file - the parity. Supports hot-swapping, good redundancy (using the term loosely) and pisspoor performance if you use the fakeraid on the motherboard (more on this later) as extra processing overhead is required to calculate the parity bits. If you have a three-drive RAID 5 setup and two hard drives fail, all your information is gone.
JBOD/Spanning
I can't for the life of me think of any reason why anyone would ever contemplate using JBOD in any setup. JBOD = "Just a bunch of disks" - data is written consecutively to each drive in the JBOD array. This has neither redundancy nor any speed benefits. As all the hard drives are treated as one large drive, if one fails all data is lost. Still, the option to use it is there.
RAID "types"
Software RAID
Can be set up and entirely administrated via the operating system without any special hardware, other than a hard drive, natch. All processing is done by the CPU instead of any dedicated hardware. Very cheap, easy-ish to set up and a bummer if the drive dies.
Fake raid
I've seen this phrase bandied about a bit - I believe it refers to motherboard RAID and the RAID setups you can get on very cheap PCI/PCI-E I/O cards which cost about a tenner and give you two or four addtional SATA ports or whatever. The difference between this and software RAID is that most of the processing has to be done by the CPU, though not actually all. The fake raid chips are then able to translate the instructions from the CPU so that whatever data needs to get written to which drive, gets written.
Hardware RAID
All processing is done on the I/O addon card. Typically these cards also have a fairly substantial amount of onboard RAM in order to help processing and throughput. Also contain alogorithms which make any read or write operations much more efficient, where possible. Cheapest I've seen is about £300. Better versions also make use of duplexing, or having two RAID controllers sharing the load. In case one controller fails, the other one can continue to function so the entire RAID array doesn't die on its arse.
Obviously, Wikipedia provides a far more in-depth description and analyses of RAID types and setups.
Testing RAID 0 and RAID 1
Hmmm. Having now established that I don't have proper RAID (hurrah for marketing types and their ability to bullshit) on the motherboard, it hardly seems worth bothering. Anyway, I did...
The system itself was dual-booting WindowsXP and Vista Business 64-bit. Both were installed onto a RAID 0 array consisting of two Maxtor 320GB which came with the system originally. Into this were placed two more hard drives, this time Hitachi 250GB Deskstars. So far, so poor. HDTach installed on WindowsXP (as there isn't a Vista version as at time of writing) and tests were done on the Deathstars; RAID being set up via BIOS and modified with nVIDIA's Control Panel depending on the test.
This is perhaps about as unscientific a set of tests as you could probably get. Still, it gave me something to do for 5 minutes.
Standalone hard drive results
The two pictures below show the results of two quick tests performed on one hard drive.
Note the "spike" in the first test, hence the retest which resulted in a smoother curve.
RAID0 test
This is the result from a quick test from a RAID0 setup.
What's interesting here is the reduction in the burst speed - this is the speed at which data can be transferred from the hard drive buffer to the interface. Possibly due to having two hard drives competing with each other for the controller's bandwidth. Sequential read speed is somewhat higher than the standalone hard drive tests.
Compare this with a similar test on the RAID0 drives containing the OS.
Average read speeds are similar, although the Maxtor RAID0 setup just beats the Hitachis - probably down to the physical characteristics of each drive.
RAID1 test
This is the result from a quick test from a RAID1 setup.
These results are very similar to the standalone drive tests in terms of sequential read, burst speed, random access times and CPU utilization. The burst speed I thought would've reflected that of the RAID0 setup but is somewhat faster. No explanation for this other than that probably the controller just reads from one hard drive in the RAID1 setup and treats it as a standalone drive - the RAID1 aspects only come into play during write operations.
Tabulated data results for (easier) comparison are below:
| Configuration | Sequential read speed (highest value) | Burst speed | Random access time | CPU Utilization | Average read |
|---|---|---|---|---|---|
| Solo Hitachi drive (test 1) | 68 MB/s | 172.4 MB/s | 12.8 ms | 0% ±2% | 55.7 MB/s |
| Solo Hitachi drive (test 2) | 68 MB/s | 176.9 MB/s | 12.6 ms | 2% ±2% | 56.1 MB/s |
| RAID0 Hitachi drive | 115 MB/s | 154.2 MB/s | 12.7 ms | 2% ±2% | 95.2 MB/s |
| RAID1 Hitachi drive | 66 MB/s | 167.8 MB/s | 12.7 ms | 2% ±2% | 56.2 MB/s |
| RAID1 Maxtor drive | 113 MB/s | 165.5 MB/s | 13.1 ms | 2% ±2% | 101.6 MB/s |
Conclusions
It's difficult actually to draw any serious conclusions from the above tests, but there are a few things that are suggested by those tests. Certainly for fakeraid there is some CPU overhead, though with today's CPUs this isn't much to worry about. For a cheap way to increase average read speeds and for high sustained sequential read speeds then RAID0 does seem to offer advantages over a standalone drive, at a cost of burst speed and obviously a more significant "cost" of data redundancy.
Home RAID1 systems are rather disappointing in terms of read speeds compared to a dedicated RAID controller but then fakeraid is basically just that. It simply provides a minor benefit of data redundancy but this should never be relied upon in place of a backup system - whether you image your entire hard drive, or in my opinion preferably, you backup just that data you consider important.
If you have bought a system with two identically-sized hard drives then experimenting with RAID0 or RAID1 could be useful if you have the time or any reason to do so. It isn't really worth the expenditure to get an extra hard drive just because your new motherboard supports fakeraid. Ultimately, it is going to be you administrating the system and you should go with whatever ultimately suits your needs. Just bear the points raised here in mind before you rush headlong into it.