I’ve been seeing a lot of IDE RAID products popping up, including motherboards that facilitate it, as well as PCI cards that do the same. So, tell me, what’s all the excitement about?
Is the purpose of RAID to simply have a large, fast, disk array, or is there some other advantage in terms of backing up data? Does it help/hurt overall system speed?
FTR, I have at least one free PCI slot, two HDs that support ATA-66 (30 and 22 GB) and one that supports ATA-100 (60 GB). What can/should I do?
(Note: I have no qualms about formatting/reinstalling stuff. Been meaning to do that to my system drive for a few weeks now anyway.)
Disk arrays are used for a combination of fast access and fault tolerance. Data can be striped across drive so that when you ask for a chunk of data you have multiplle drives each getting a portion of your data. Data can be redundantly recorded by mirroring drives. If one dies you keep chugging along. RAID-5 IIRC stripes across a number of drives but retains one to hold checksums. If one drive dies the data can be rebuilt with a minor delay but ideally with no loss of information.
I don’t think you’ll see benefit in an arrary for a single user system.
IDE RAID sounds good but in practice gets you little to nothing.
RAID-1 (Mirroring) will give you fault tolerance but at the expense of one lost drive (i.e. two 30GB drives mirrored leaves you with 30GB of data storage…not 60GB). If your main drive dies your system dies with it usually. You need to promote the mirrored drive to master status and then you’re back in business.
RAID 0 will give you striping across two or more drive but with NO fault tolerance. Two 30GB drives in RAID-0 will get you 60GB of data storage.
I’ve never seen RAID-5 IDE but RAID-5 SCSI gets you fault tolerance such that if one drive dies the system keeps working anyway. Plug in a new drive and it gets rebuilt automatically. RAID-4 maintains a drive specifically for the checksum. RAID-5 stripes the checksum across all drives. I can’t remember why but RAID-5 is a bit better than RAID-4 (you almost never see RAID-4 anymore). You only lose a fraction of your drive space to RAID-5. If you had 5 drives, for example, you lose about 20% of your drive space to the fault tolerance (as opposed to 50% in a mirror setup).
Theoretically RAID systems are faster. In a mirror (duplex actually…subtle difference) system READs from the drive can happen on two drives at once. Two heads reading are faster than one. WRITEs however happen only to the Master drive and it then tells the mirror drive to mirror what was written so no speed benefit is gained there.
In a RAID-0 system you can READ and WRITE to both drives at once so again there should be a speed benefit.
I’m having a hard time fiding an article that benchmarked IDE-RAID performance and it was somewhat disappointing. Not bad but not really worth the price of admission either.
Too bad too since I own an IDE-RAID capable motherboard although I do not have RAID setup. I guess I figured maybe someday via driver updates or something a benefit could be realized (and I do get 4 IDE hookups instead of two which is nice).
Jeff_42 did an excellent job of summarizing the RAID proposition for single user setups. Here is a site I can recommend that will tell you more than you ever wanted to know about hard drives, IDE raid etc etc. Also has reviews of actual IDE raid setups and measured performance.
I have been running Raid-0 for about 4 months, and I’ll tell ya, there is no better hard drive configuration. It is way faster and cheaper than SCSI, and since it uses regular IDE drives, the possibilities are endless. I have 2 IBM 40gig Deskstars, and these things are smoking…if you can go with Raid, do so
There are also high performance IDE RAID subsystems available from other parties (Doing a Yahoo search on IDE RAID will get you some info), but they generally use an outboard RAID controller connected to a PCI SCSI-3 interface on the motherboard.
Low-cost IDE RAID should be implemented for reliability, not performance, as IDE RAID performance is actually lower than that of a single IDE disk (unfortunately, I don’t remember where I read this, but it was probably on StorageReview.com or Tom’s.
I’ve yet to check out the links provided (thanks guys!), but I had a question about the “no fault tolerance” Jeff noted about RAID 0. Does this mean, as I think I read somewhere, that if one drive dies that the data stored across the two drives is effectively hosed?
Sorry it took so long for a response. I hope this is still helpful KK.
Yes. If one drive goes in a RAID-0 then the whole thing is hosed. There are companies that can recover data from smashed defective drives but its expensive and only worth doing if the data on the drive is very valuable.
Take this with a grain of salt however. RAID-0 is not really any worse than a single drive. If a single drive goes you’re in the same boat. Of course you’re rolling the dice for a breakdown twice now instead of just once (two drives) so I guess the chances of losing everything are actually increased in a RAID-0 setup.
RAID-1 is mirroring (or duplexing) in which the contents of one drive are copied (mirrored) onto a second drive. If one drive dies you still have the data on the second drive. If the master drive dies the system will crash. It is then necessary to manually promote the slave drive to master status and you’re back in business. Also, you effectively lose one whole drive for the fault tolerance so this is a somewhat expensive option.
RAID-2, 3 and 4 aren’t done anymore (at least I never see them).
RAID-5 gives the best performance and fault tolerance. Data is striped across all drives with a checksum for fault tolerance. One drive can die and the system, using the checksum data, will keep running (albeit at a somewhat slower overall speed). If the bad drive is replaced the lost data is automatically rebuilt on the drive. If the drives are hot swappable (i.e. you can remove a drive without shutting the system down) the drive can be replaced without any interruption in service. In addition performance is increased since you have X number of drives reading and writing data simultaneously.
The only downside to RAID-5 is it’s expensive. The RAID controllers for this aren’t cheap and you need a minimum of three drives to make it work. Having a hot pluggable capable system only makes things more expensive. Usually only server class machines require this level of protection and performance but it is very slick stuff.
As a DBA (database administrator), I work with RAID systems quite a bit. One advantage of RAID-1 is that you can theoretically have faster reads (at least in a multi-user environment) because while one drive is serving user A, the other drive can be serving user B. Or drive A can be seeking while drive B is reading.
I saw RAID-3 recently on a RAID manufacturer’s website. There was an advantage vs. RAID 5, but I don’t recall what it was, and I’m too lazy to look it up. Actually, the site I reference below says that write performance is better than RAID 5.
Actually, there’s another disadvantage to RAID-5 - it’s slower for writing than either a single disk or RAID 1.
You left out two other common setups: RAID 0+1 and RAID 10. RAID 0+1 is a pair of RAID 0 arrays that are mirrored. RAID 10 is a RAID 0 array, where each element in the array is a pair of mirrored disks. The cost of implementing either of these is on par with RAID 1 (i.e. you need twice as many disks for a given capacity as you would for RAID 0 or JBOD [Just a Bunch Of Disks]).
As a DBA, if cost is an issue I’ll use RAID 5 for the data storage (where read performance makes more of an impact) and RAID 1 for the (transaction) logs, where fast write performance is important. If cost is not as important, I’ll use RAID 10. I’m actually still a little unclear on the performance & reliability differences between 0+1 and 10 (even after reading this site), but either is faster than 5.