What is Digg?96 Comments
- borninda818, on 10/11/2007, -0/+114http://www.epidauros.be/raid.jpg
this picture may help those still having a hard time understanding. - shifty2, on 10/11/2007, -8/+62this is a good article explaining the different type of RAID configurations as well as the pro's and con's of each. It is straight forward and to the point, key terminology is well defined and simple to understand and *****, the author made simple to understand pictures of how data is distributed and stored for each RAID type. the pictures alone negate some reading.
only a house fly w/ down syndrome would have a problem with the length of this article. - Error601, on 10/11/2007, -2/+37So you can lose 1TB of data in one shot?
- Error601, on 10/11/2007, -0/+26Did someone say it was? A good backup schedule is never a substitute for redundant disk storage either. You are guaranteed to lose disks over time and it should be a zero impact event.
- ExSlashdotter, on 10/11/2007, -3/+28RAID 5 is never a substitute for a good backup schedule.
- tempusrob, on 10/11/2007, -1/+20"RAID 5 is never a substitute for a good backup schedule."
Indeed, the "I have RAID, I don't need to do backups" attitude is probably the biggest myth regarding RAID. RAID is *irrelevant* when it comes to backup, and shouldn't even be used in the same sentence. RAID is for redundancy and continuous availability. It won't help you recover deleted or corrupted files, it won't help you roll back to a previous version of a file, and it can't be taken offsite for archival purposes. - dbre2, on 10/11/2007, -2/+17Isn't that where all the 1's are stored on one hard drive and all the 0's are on the other?
- SavageBlackCat, on 10/11/2007, -4/+19Written by someone with a better understanding: http://www.acnc.com/04_01_00.html
- Namco, on 10/11/2007, -1/+15I've had this one on the wall of my cube for over 2 years. It's classic.
- Nougat, on 10/11/2007, -0/+14Dugg for explaining RAID3 and RAID4. I always wondered where those went.
- Bakkster, on 10/11/2007, -3/+16@sinembarg0
In theory, yes. In actuality, no.
The problem is that hard drives have a seek time that is the time for the read head to find the right track plus the time for the data to rotate under the read head. This is usually 0.5s on modern hard drives. While RAID1 can do parallel accesses to halve the access time, the seek time is NOT reduced. Seek time is the dominating factor in disc reads, especially in the age of 3Gb/s SATA drives.
So for a file that takes 0.6s to read on a single disc, on a RAID1 it will still take 0.55s. Hardly a significant increase. Add that a fragmented file access has a seek time for each fragment, and you see that under most circumstances RAID1 does very little for access time. - understudy, on 10/11/2007, -3/+14
I've always just sprayed it and the roaches die.
Wink.
_ - paulmdx, on 10/11/2007, -1/+9Sorry, I should have qualified that with: for databases with a medium/high amount of data writing.
- championchap, on 10/11/2007, -4/+12I'm digging you down too, it wasn't funny.
- Wyzard, on 10/11/2007, -1/+8@Bakkster:
"[Seek time] is usually 0.5s on modern hard drives."
Actually, it's about two orders of magnitude less than that. The datasheet for the Seagate 7200.10 family of drives, for example, reports an average latency of 4.16 milliseconds. - catalysis, on 10/11/2007, -1/+8FTA: "RAID0 is one of the most common RAID levels used in desktop systems for increased performance as today it is integrated on nearly every new motherboard."
Now you know. - JavaApe, on 10/11/2007, -0/+7That's what /bin/true and /bin/false are for. I store all my ones in /bin/true, and sure enough, I can read them back any time I want. I use a similar system to store all my zeros in /bin/false. The only problem is figuring out in what order I need to read my data back -- but I think a push-down stack may be just what I need. Isn't it great to be a system administrator?
- shifty2, on 10/11/2007, -4/+11^^ that's awesome!! i wish i had that pic like 6 months ago when i had to explain to my CTO (ya, i dont know how he got the job either) why we need at least a RAID 1 setup for our DATA SERVER!!
- Anoobis, on 10/11/2007, -1/+8But then it says:
From a performance point of view however, RAID doesn’t make much of an argument for desktop use. Sure, you may save a second when loading a game, but the same time could be saved by upgrading your CPU or memory. Fact of the matter is, unless you’re doing some pretty hardcore graphics, audio or video work, the performance potential of RAID doesn’t mean much. - cbrunet, on 10/11/2007, -2/+8And when your whole life is deleted cause that one drive fails?
Sure would have been nice to have that RAID 5'ved, wouldn't it? - Error601, on 10/11/2007, -0/+6You still have to know mirror from raidz. I've been running it on my solaris boxes for a while and it's very slick.
- Bakkster, on 10/11/2007, -0/+6@Wyzard
Thanks for the catch, my units were a little off ;)
Regardless, the seek time is still the dominating factor for most reads, excepting large defragmented files, so RAID1's performance increase is minimal. - kbull, on 10/11/2007, -1/+6Anyone who chooses to deploy a server in their business without RAID deserves to lose their data. Now, with people building multi-terabyte media servers at home you'd be crazy to not have some redundancy there as well. Also, RAID and a regular backup schedule are NOT mutually exclusive.
My rule... No RAID = No server for you! - Wyzard, on 10/11/2007, -0/+5@smackywentz:
"Dual-channel", a feature of many current enthusiast motherboards, is essentially RAID-0 for RAM: no redundancy, but striping for better performance. On a motherboard that supports it, it's usually used automatically if the RAM is installed in a configuration which supports it (typically identical DIMMs in adjacent slots).
For each 128-bit chunk of data to be stored in memory, 64 bits of it is stored in the first DIMM and the other 64 bits is stored in the second DIMM, and the "dual channels", one to each DIMM, mean that the two chunks can be read and written simultaneously as a single unit.
http://en.wikipedia.org/wiki/Dual-channel_architecture - maehem, on 10/11/2007, -1/+6Switch to ZFS and stop worrying how RAID works.
"ZFS. So simple even a cave man can do it!" - theatrus, on 10/11/2007, -1/+6Yes, but you need some seriously fast drives to be able to exceed 70MB/s from the inside edge of your disk platters, so the point is mute. The only time you will ever see the SATA or PATA interface link saturated is when reading from the drive buffer, all whopping 16MB of it, or you use a very high speed solid state disk :).
And I never consider RAID0 to be superior unless its a 0+1 setup. - cbrunet, on 10/11/2007, -2/+7It wasn't funny. Did you not get the memo that WoW is mainstream now and not cool for geeks anymore?
- dbalaski, on 10/11/2007, -0/+4Try looking at a secondary Raid Controller Card:
Such as -- just a few examples
SuperTrak EX12350 (12 Port Raid-6)
SuperTrak EX4350 (4 port Raid-6)
FastTrak TX4310 (4 Port Raid-5 ) - Error601, on 10/11/2007, -0/+4Get a forth 160GB drive and run mirrors. Three disk RAID5 is a lot of disadvantage for a not much advantage.
- TVarmy, on 10/11/2007, -0/+4I'm building a PC that I intended to have RAID 5, but I don't have a RAID card, and I am instead using NVRAID with a dual core AMD processor. The PC will mostly be used for office tasks like word processing, email, etc., with occasional use of a database/scheduling program (ACT, if you've heard of it). Is this a mistake, and should I instead just switch to a RAID 1 setup? Redundancy is a higher priority, but RAID 5 sounded like a good deal before I heard about how intensive on board RAID can be.
- theatrus, on 10/11/2007, -1/+5@TempusRob
Totally agree.
Except some of those cases are solved already. Which is what is awesome about a filesystem such as ZFS. Solves the RAID 5/6 write hole problem in software, and has very very good support for snapshots (which are constant time to create, and easy to view and manage). A popular use case on fileservers is to make snapshots on the minute/5minute/hour/4 hour/day/week schedule, which can save you tons of time on restoring user deleted files from backup media.Plus the filesystem is heavily checksummed, so even disk controllers scrambling bits isn't going to corrupt data (or at least it will be detected well before half your disk is random bits).
Once again, no subsitute for a good offsite backup schedule, but modern file systems like ZFS can make bad IT tasks such as restoring from backups because someone deleted their word document much easier.
The offsite backups are there for when all your drives cook off due to a melted server, or catch fire. - cbrunet, on 10/11/2007, -0/+4Depends what you mean by greater. Do you mean faster? Then yes, the SATA will be faster. Do you mean better at disaster recovery? Cause then you would be wrong, and the RAID0+1 or RAID5 would be your best best.
- merelyjim, on 10/11/2007, -0/+4Me again. Just wanted to say "thanks" for the input...
- nonymous666, on 10/11/2007, -0/+3In most RAID controllers (i.e., the cheap on-board ones that motherboards have, not the more expensive (smarter) external PCI controllers), read access involves only reading from one drive. The second drive isn't accessed during reads, only during writes. Yes, some RAID controllers do parallel reads from both drives, but you're going to have to pay for them.
- Philluminati, on 10/11/2007, -0/+3lol, your parity check could be 111111 XOR 000000 and you just dump the output on /bin/false (1) again!
- TVarmy, on 10/11/2007, -0/+3Forgot to mention, the drives are 3 160 GB Seagate SATA drives.
- dbalaski, on 10/11/2007, -1/+4No -- Raid-5 can work great for DBs with medium-high write volume. (I do this all the time, one of my largest databases is 1+TB database with 6000 active users -- we are using EMC symetrics arrays on Sun E25K machines using fibre channel interconnects )
Things I would add:
Use Hardware Raid Controllers -- a raid capable controller with a decent amount of non-volatile cache.
Software raid solutions (where OS handles the striping/mirroring etc) have too much a performance impact on the systems.
Agreed -- RAID-5 (or any raid) isn't a substitute for a backup. if you think it is a substitute, your in the wrong business and you don't think problems thru... (e.g. won't help you recover data that has been deleted/changed, such as file or data inside a database table ) - Error601, on 10/11/2007, -0/+3Those EMC boxes aren't your average RAID system. I tested some for one job and could find no performance hits for the redundancy. Of course those units cost more than my house at the time.
- Wyzard, on 10/11/2007, -0/+3@JavaApe:
You got it backward. :-) /bin/true returns zero because that's the code for success, and /bin/false returns 1 because nonzero indicates failure. - Anoobis, on 10/11/2007, -1/+3newbie question:
SATA3 RAID 0 > IDE133 RAID 0+1 or 5 ?
as isn't sata 3 much quicker than 133? - MikeOSX, on 10/11/2007, -0/+2It is possible to do. However, note that if you plan on constantly plugging and unplugging your drives, you face a great chance of "breaking" the raid. You may lose all your data, depending on the raid configuration. If you are using mirroring, you may be able to rebuild the raid, but that usually takes forever to do. Raids are more ideal for desktop usage, where they will be sitting in the same spot all the time.
- KoZo, on 10/11/2007, -0/+2I have a pretty good understanding of RAID, having created a Linux RAID monitoring system for x86 servers, specifically Intel's offering. Though RAID devices are pretty expensive, there are poor man's way to create now. If I have money I will shell for one.
- Nadare, on 10/11/2007, -0/+2I've purchased some highpoint raid 5 capable card, and I can tell you from my experience that it too was intensive on the cpu ( when compared to just a single drive), maybe it needs a 'xor engine' or whatever to lower its cpu usage, but I would just buy two new larger drives (500gig) and use the on board raid 1.
With your 3 or 4 disk raid 5, your performance will suffer until you replace the drive and the data is rebuilt, plus in reality you might want to keep a spare drive handy, because if during your downtime of one drive failing you have zero reliability - that is unless you're willing to send the drives to some specialty place to recover the data.
Raid 1 is very simple, so it shouldn't stress the system. - Error601, on 10/11/2007, -3/+5Why? Works very well for high capacity, heavy read, low write data warehouses.
- theatrus, on 10/11/2007, -1/+3On a modern machine, unless you need every spare CPU cycle, software RAID-5 is not a bad choice. Just watch out for random issues with the NVRAID controller system... Some motherboards like ejecting disks from the array at random for no reason.
I can't speak for the SuperTrak controllers, but the FastTrak controllers from Promise fall into the realm of FakeRAID. They do not do actual hardware RAID once the system is no longer using BIOS calls to read from the disk (those first 5 seconds of boot), instead doing it in a software driver, gaining you no speed boost for dedicated hardware. I don't think software RAID is bad at all, but beware of using non-standard software RAID (so Linux md driver, ZFS = good, random other controller which you won't be able to buy in a year when your old one catches fire = bad).
If you must have dedicated hardware RAID on SATA, look at 3ware and Areca. - Ramble, on 10/11/2007, -0/+2Don't expect the performance to be greatly higher (chances are it'll be lower).
- merelyjim, on 10/11/2007, -0/+1Anyone know if you can set up a RAID array using external USB hard drives? Be nice to set something up like that on my laptop for security, and then unplug and go.
And, yeah, I'd like to do it short of the Drobo... - Anoobis, on 10/11/2007, -1/+2argh i meant
SATA3 RAID **1** > IDE133 RAID 0+1 or 5 ?
as this will give speed and data recovery in case of a drive death?
but some interesting points there thanks v much - blackoper, on 10/11/2007, -0/+1raid 0 that is raid 1'd for data protection, or a very good raid 5 solution. I'd have both use 15K rpm drives, preferably of the scsi u320 variety or fiber channel if it's enterprise/business class data. If I had to use a sata solution I'd get the sata 2 raptors.
- blackoper, on 10/11/2007, -0/+1One thing I have to disagree with in the article is that Linux kernels seem to have no problem keeping up with hardware controllers at least with raid 5 and video/tv/archive/fileserver use with only a handful of concurrent connections. Speeds were roughly the same and very little overhead. I have two raid 5's and a raid 0 array on my server. 12 active drives with two hotspares.
Also I'd put a dedicated openfiler server with raid 5 going through pci-express interfaces (port multipliers, etc) up against a much more expensive hardware/commercial solution all day long. -
Show 51 - 93 of 93 discussions
© Digg Inc. 2009
— Content created and posted by Digg users is 