I <3 Steve McConnell*
Coding Horror
programming and human factors
by Jeff Atwood

July 07, 2005

Desktop RAID: Oversold?

I've seen a number of hardware-oriented developers talk about setting up striped RAID arrays on their personal desktops. It does seem like a reasonable idea, given the current strong trend towards "doubling up" on hardware to leverage performance benefits from parallelism in various forms-- dual core CPUs, dual channel DDR, dual graphics cards in SLI, and dual hard drives in RAID 0.

However, except for dual channel memory, none of these parallel hardware approaches are clear across-the-board performance wins. They can be twice as fast in very specific circumstances, but as a general rule they're just somewhat faster, and not without adding significant costs and even risks of their own.

RAID 0 (aka striping) on the desktop, is especially dubious in my opinion. Just because current motherboard chipsets now make it easy and (relatively) inexpensive to do this doesn't mean that you should:

  1. RAID 0 literally doubles your chance of drive failure. All it takes is a single drive failure on either drive to render your striped array completely and utterly unrecoverable. A "normal" single drive failure is at least theoretically recoverable, albeit expensive. How expensive? If you have to ask, you probably can't afford it.
  2. Of all the hardware failures you can have, drive failure is by far the most traumatic. Losing a CPU, video card, or even motherboard, means you just go out and buy another one. Losing a drive means you've probably lost your critical data, unless you have a good backup regimen in place. And 99% of us don't. Kudos to the 1% of you reading this who do.
  3. The performance benefits of RAID 0 aren't that compelling for typical desktop usage scenarios.

So why would you go to the trouble of building a RAID 0 array if it's more expensive, more complex, prone to failure, and not all that much faster? Short answer: you probably shouldn't, unless you know you have a specific scenario that justifies this setup. There are a number of articles on the web that debunk the myth that RAID 0 is a universal performance improvement for a typical desktop PC:

All these articles are variations on the same theme; the AnandTech article summarizes nicely:

If you haven't gotten the hint by now, we'll spell it out for you: there is no place and no need for a RAID-0 array on a desktop computer. The real world performance increases are negligible at best and the reduction in reliability, thanks to a halving of the mean time between failure, makes RAID-0 far from worth it on the desktop.

There are some exceptions, especially if you are running a particular application that itself benefits considerably from a striped array, and obviously, our comments do not apply to server-class IO of any sort. But for the vast majority of desktop users and gamers alike, save your money and stay away from RAID-0.

The original hard drive benchmark review website, storagereview.com, offers this guidance:

The enthusiasm of the power user community combined with the marketing apparatus of firms catering to such crowds has led to an extraordinarily erroneous belief that striping data across two or more drives yields significant performance benefits for the majority of non-server uses. This could not be farther from the truth! Non-server use, even in heavy multitasking situations, generates lower-depth, highly-localized access patterns where read-ahead and write-back strategies dominate. Theory has told those willing to listen that striping does not yield significant performance benefits. Some time ago, a controlled, empirical test backed what theory suggested. Doubts still lingered- irrationally, many believed that results would somehow be different if the array was based off of an SATA or SCSI interface. As shown above, the results are the same. Save your time, money and data- leave RAID for the servers!

If you have your heart set on RAID 0, go for it. It won't hurt performance. But be sure you're making an informed decision. There's a lot to be said for simplicity.

Posted by Jeff Atwood    View blog reactions

 

« Stupid Command Prompt Tricks How to be Successful, Happy, Fulfilled, and Drive a Totally Hot Car »

 

Comments

I just built a computer about 2 months ago, and I considered doing RAID 0, but I decided against it for the reasons you've cited here.

I think a close parallel right now also, is nVidia's SLi and ATI's Crossfire technologies. The performance advantage gained from daisy-chaining two video cards together is no where near worth the cost involved. The primary difference between this and RAID 0, is that the dual video card setup is likely to improve as more games begin to support it, whereas the downsides of RAID 0 are unlikely to go away anytime soon.

All in all, it really depends on what your needs are. If you absolutely must have a great deal of speed, and you're not concerned about data integrity, go ahead and RAID it. If you'd like a boost in speed, but are concerned about integrity, just step up to a drive that spins at 10k rpm. If you're ultimate concern is data integrity, then you might consider RAID 1. Most motherboards that support a RAID setup at all will be able to do it as well as RAID 0, and you're much less likely to lose any of that data.

Something I don't know much about is RAID 0+1. Just by the suffix, it sounds like it would require 4 disks, but I'm just talking out my ass on that.

Marty Thompson on July 8, 2005 02:27 PM

RAID 0+1 is exactly that, from my understanding. 2 striped disks, mirrored onto 2 other striped disks.

ALL HOME USER PC's should be RAID1 or better. HOME USERS HAVE _*NO*_ BACKUP STRATEGIES to speak of! When their cheap @ss HDD fails, too many friends have said "What can I do to recover my photos of my trip to "X" where "X" is somewhere they will never visit again. I suggest a drive recovery company ($600-$2000) And they balk. Sucks to not have redundant data. How important _IS_ your data?

Brian Hampson on July 8, 2005 02:59 PM

I have a couple of suggestions.
1) If you're going the multiple drive route, why not partition the data according to how it's going to be accessed. I usually put my (barely used)swapfile and database files on my second drive, with the OS and projects on the first drive. That way I can have I/O going on both sets of devices more or less simultaniously.
2) External drives have gotten cheap. I picked up a Maxtor 300 GB drive for ~ $220. You can back up to it reasonably fast, and it won't die if your MB fries your internal drives.
3) How are you going to restore? I have two installs of XP on seperate drives. If one goes, I can replace the drive, boot the other copy of XP and restore the first one. Theoretically.

Mike Swaim on July 8, 2005 03:13 PM

Thanks for the good thoughts, and the links. Last December I bought and assembled a new home-brewed gaming rig. I was shooting for a combination of fast and quiet. (My last desktop was obnoxiously loud.)

One additional consideration to think about, therefore, is noise. I thought that I could get a good noise-performance ratio by going with two quiet SATA hard drives in a RAID 0 configuration.

However, since they're both being accessed at the same time, they actually seem to increase the noise. (Especially when the "clank" over to a new sector/track/whatever in unison.) They also increase the case temperature, to some extent, by having multiple drives.

I had assumed that I had at least obtained a significant performance increase. From the AnandTech article, I see that that's probably illusory.

Robert Jacobson on July 8, 2005 03:53 PM

When I built my last system, I did go for RAID 0 across two SATA drives. I really like having one large disk, and the speed is good. I haven't compared it to a single drive (other than my older slower system), but it seems fast to me.

I'm aware of the reliabilty issue, and was making backups regularly for a while, but have started slacking off.

Maybe I should use them as seperate drives on my next OS reinstall.

David Crowell on July 8, 2005 03:55 PM

I had a motherboard with a RAID 0 set die on me. Sucked because I couldn't read my data off without buying another MB with the EXACT same controller, and that was not easy to find a year and half later in the computer market. Luckily I found one, and my new computer is SATA RAID 0 w/ WD Raptor. I really like it, but to offset the worry of failure I bought an under $200 120 gig Maxtor external and backup to it every night so I could at least read the data on the other machine. Combine that with VCS on a server and I'm pretty comfortable that most of my important data is safe.

Ryan McGinty on July 8, 2005 04:13 PM

Nowadays I think RAID1 is the bare minimum any computer should have. With hard drives currently it's not IF it's going to fail, it's WHEN is it going to fail (usually 1 month after the warranty run out :-)

RAID5 is probably the best bang for your buck, but not many home controllers support it (imho). I can't see why anyone would use RAID 0+1 instead of RAID5.

Brian Mowrey on July 8, 2005 04:25 PM

If somebody is looking for a good backup software, you should check out Acronis True Image. Truely great software, which makes it possible that I can backup the last 3 weeks (backup every day) of my 200GB data onto a 250GB drive.

Hermann Klinke on July 8, 2005 04:33 PM

> If you'd like a boost in speed, but are concerned about integrity, just step up to a drive that spins at 10k rpm

This is definitely true and good advice. After looking at these benchmarks, I recently upgraded to a 74gb 10k RPM Raptor drive as my system/OS drive, and relegated my existing 250gb SATA to data duties.

The performance benefit of the 10k raptors is *definitely* measurable in boot and load times. And it's noticeable. Surprisingly, it's not that loud either. There's virtually no idle whine at all, and seeks aren't that loud. I was quite impressed!

> Nowadays I think RAID1 is the bare minimum any computer should have.

Right, but as Ryan pointed out, you better have the exact same mobo/chipset to restore that data. I think *software* RAID1 is a better solution, ultimately, because it's more flexible and can be deployed/restored on any mobo. It's too bad that you can't do RAID1 in a convenient way using XP Pro. I believe it disallows mirroring on the boot volume which makes it almost useless..

Jeff Atwood on July 8, 2005 05:47 PM

Who out there finds a good SATA drive too slow? I switched to SATA from U160 SCSI and never looked back. No need for RAID on desktop... SATA is the fastest and most economical solution for the desktop out there. (A lot quieter than fast SCSIs too.)

Gene on July 12, 2005 07:30 PM

>Right, but as Ryan pointed out, you better have the exact same mobo/chipset to restore that data. I think *software* RAID1 is a better solution, ultimately, because it's more flexible and can be deployed/restored on any mobo. It's too bad that you can't do RAID1 in a convenient way using XP Pro. I believe it disallows mirroring on the boot volume which makes it almost useless..

I've found that disks that are RAID1 on the MOBO quite happily work as standalone, and as the extension of that, they work on a new MOBO. Then, if I need another MOBO, I just mirror based on the functional disk that I have (assuming one of the two fried with the MOBO), and I'm back in business. Done it. It works.

Brian Hampson on July 13, 2005 04:04 PM

> I've found that disks that are RAID1 on the MOBO quite happily work as standalone,

Oh yeah, you're right, I wasn't thinking that through. It's RAID 0, striping, that is dependent on the implementation. RAID 1, mirroring, should be a simple, perfect 1:1 copy of the data on each drive.

> I switched to SATA from U160 SCSI and never looked back

I can definitely recommend the 74gb 10,000rpm SATA Raptor drive as a system/boot drive. The difference in speed over the 7,200rpm drives is very noticeable in actual use. Bootup and load times are significantly reduced. And it's not particularly noisy, either!

Jeff Atwood on July 13, 2005 05:35 PM

I have installed two Maxtor Sata 80MB drives hooked up as raid 0. I also have another hard drive hooked up as a standard IDE drive used for backup. My Sata drives have crashed three times. The first time was using the Scan Disk programs from Microsoft, The next time it was using the Defrag program from microsoft. The last time was loading the Norton System Works Go Back program. All of these required reformatting the hard drive. The hard drives are perfect--just programs fouling up the boot record and registery. Windows will not boot. Windows cannot repair. Windows does not recognize the drive.

I also have used 4 different Video editing and Authoring programs. Some have trouble with my drive. Some do not read in the full clips. Others take 1 hour per minute of video to render however it is successful.

I learned the hard drive way!

John McGinnis on December 10, 2005 01:33 AM

> I can't see why anyone would use RAID 0+1 instead of RAID5.
There's pros and cons to both setups. For personal use, I would choose 0+1 over Raid5 because the write times are much faster. Raid5 has a major hit in write performance due to the parity involved. In many office situations, data is more often read than written (& typically smaller files like Docs), so raid5 is fine. The catch with 0+1 is you only get half the drive space available versus raid5's "Total Drives - 1" amount (4x200GB=800GB, but only 600GB is available ).

Adam on January 21, 2006 02:38 PM

I'm reading this thread 'cause my warranty on two Raptors in RAID0 is over now, and though it's been the fastest desktop performance I've ever tried (under FreeBSD and 1GB of RAM) on a desktop, I'm a bit worried - no more warranty, but the noise of "clicks and clacks" is getting worse every day now :D
Do the disks really die soon after it starts happening? :)

Isnogood on March 23, 2006 07:26 PM

If your disks are making unusual noises, BACK UP YOUR DATA IMMEDIATELY. That's never, ever good!

And this goes double if you're on RAID 0 (striping). That means your chance of failure doubles *and* there's no easy way to recover data from a striped drive!

Jeff Atwood on March 23, 2006 08:02 PM

Let me start off by saying ANY form of RAID is not meant for the typical home user. Do you seriously think Uncle Joe or Grandma Lou-Anne will even know what it is? You should have some clue of what you are doing when you decide you want to build a RAID array. Those who know what they are doing, know the concequences.

With modern processors (the Dual-cores, AMD X2/FX-series, Hyperthreaders, etc.) the bottleneck is not so much the system bus or the memory speed anymore - it is, you guessed it, the hard drive. RAID 0 has a significant improvement depending on how new or advanced your system is. The new SATA-II drives (3.0GB/s) in a RAID array will run laps around just about any drive out there today.

I noticed other people saying just to buy a 10,000RPM drive. Wrong! Not only will the difference in speed be minimal compared to a RAID 0 array, but the real delay in hard drives is the time it takes the drive to adjust to the platters. Not to mention, the faster drives create more heat, thereby not decresing your chances of failure that much either (perhaps even increasing them).

Now, I'm not denying the chances of failure are great. I'm not suggesting that at all. But going back to my first point: those who setup RAID should know what they are doing. ANY GOOD COMPUTER USER KNOWS TO BACKUP THEIR FILES! That doesn't always mean on another drive or partition (I use an external USB hard drive), but also DVDs or CDs as well. I use Tweak UI to specify to automatically put all of my personal stuff in the "My Documents" folder onto the external drive. WD also makes an external USB drive with automatic backup feature. It's like building a street car... how much are you willing to spend to go fast.

Greg on May 13, 2006 07:06 AM

> I noticed other people saying just to buy a 10,000RPM drive. Wrong! Not only will the difference in speed be minimal compared to a RAID 0 array, but the real delay in hard drives is the time it takes the drive to adjust to the platters. Not to mention, the faster drives create more heat, thereby not decresing your chances of failure that much either (perhaps even increasing them).

You're making a lot of claims without providing any evidence to back them up.

*None* of the benchmark or reliability data I've seen supports what you are saying. In fact, you're dangerously close to misinformation.

If you like desktop RAID, fine, but the data tells me that A) there's no meaningful performance improvement for the vast majority of apps, and B) it doubles your chance of drive failure.

Now, on a server, it's a different story.

The future of desktop hard drives is probably hybridized flash memory/magnetic platter drives.

Jeff Atwood on May 14, 2006 01:50 PM

Ok so I run Raid0 on two WD SATA drives. Since I first started using it I have had 5 crashes. These 5 crashes resulted in the following:

- 3 HDD drive replacements
- 4 motherboard replacements including switching from AMD to Intel
- Countless OS re-installs
- Countless re-activation of Raid array
- PC spent 1.5 months in a shop while we tried to source identical motherboard

PC is used for games. Not really sure how much of a performance increase RAID0 is but right now its 12:39am. I have been trying to get my PC up and running...after it crashed 1 week after getting it back from the 1.5 month stay at the PC shop. I have had enough. Kicked the PC across the room the last time this happen (not scientific and no damage to PC - 15kg thermaltake case with all the bits and pieces). Left me with a sliced toe but the satisfaction was enormous)...wife walks in and says...."baby, just go buy what you need for Raid5....". Got to love her. So I am going to go back to standard SATA and use my 300 SATA 2 drive for games. I worked for Microsoft for 4 years....resigned to sell motorcycles because I hate working with PC's...seems to be the right decision.

elton - still a PC junkie
1 x HP notebook
1 x Acer notebook
1 x LG notebook
1 x Media Center PC
1 x Storage server
1 x AMD games PC
1 x Intel games PC

Elton Wilmot on July 26, 2006 01:49 PM

I've got a pair of 120Gb SATA disks hooked up to my desktop PC as RAID0, and I have to say that the performance improvement over 2 separate volumes is noticeable.

I was doing some scalability testing on a large database not long ago, and the RAID0 configuration was about twice as fast as the separate disk configuration, on roughly similar hardware.

Of course, with DB programming, there's always the possibility of manually splitting the database across multiple spindles (data files vs. log files vs. tempdb), but I've noticed the speed in other scenarios as well.

On the other hand, I _do_ have a reasonable backup regime in place: All of my data files are stored on a networked server with RAID5, which is periodically (although not as often as I should) backed up to external (USB 2.0) disks.

Roger Lipscombe on October 31, 2006 01:59 AM

I've read the benchmark articles but none seem to mention compilation performance. Has anyone done any benchmarks comparing compilation times for (say) a 10k RPM raptor versus a pair of them in RAID 0?

I'm looking to upgrade the desktop PCs for the developers at work and some guys have been pushing hard that RAID-0 is a big performance win for compiling. Is it?

Ben Robbins on October 31, 2006 05:57 AM

> I'm looking to upgrade the desktop PCs for the developers at work and some guys have been pushing hard that RAID-0 is a big performance win for compiling. Is it?

Highly unlikely. Just get them 10,000 rpm Western Digital raptors for best performance, and forget about RAID.

Jeff Atwood on October 31, 2006 06:39 AM

The question I am left with reading this thread is why is backup so avoided by everyone, including "professionals"? It is almost as if anything at all is better than backing up - even spending the same money on RAID1 as a 'pretend' backup.

It has taken me so long to perfect my PC, and the work I do on it represents such a large chunk of my life, and in some cases I have clients depending on it, I don't think I could live without 2 copies of backup (one stored offsite) and a proven (e.g. tested) recovery plan.

Every time an IT professional suggests to me that I would be better off using RAID1 instead, I ask them what good will it do me if I accidentally delete a file, have my house burn down or my PC stolen in a burgulary? The usual response is a blank look, a long pause, and the answer "that never occurred to me".

Paul Coddington on November 1, 2006 10:07 PM

Remember, the RAID advice I've provided is for desktops. For the *server* situation described by Roger Lipscombe:

>I was doing some scalability testing on a large database not long ago, and the RAID0 configuration was about twice as fast as the separate disk configuration

.. the answer may very well be different.

Jeff Atwood on November 2, 2006 07:01 AM

So, if I want to build a gaming PC with 2 7,200 RPM HDDs in a RAID-1 setup (which should almost double read speed), is there any point to getting a 10,000 RPM Raptor for the OS drive?

Bob on November 7, 2006 02:30 PM

I back up my data to 2 ide's. One I leave plugged in and the other I store in a box as a secondary. My Satas are in raid 0.

With RAID 5 and 1 you are only protected against hd failures. You still need a back up the data to another HD or tape drive. That cost money. RAID or no RAID.

If a primary hd goes bad you have to replace and reinstall either way. In fact, if you didnt have the money you could just take the raid 0 out and you have another hard drive already.

If your dumb enough not to have a backup then obviously raid is not for you. This article assumes 99 percent of people are that dumb.

And to the above poster.. Raid 1 does not double your read speed. It is comparable to a single drive.


brett on November 10, 2006 10:47 AM

I've been using RAID 0 for my two home desktops and it's been a great time saver. I use Photoshop heavily and process thousands of files, so the speed of opening and saving files is worth every second of savings.

Daniel Payne on November 26, 2006 11:59 AM

Jeff. Perhaps its time to replace this outdated blog with one more in tune with todays reality. First your correct when you say Grandma @ Grandpa who play cards online or access email on occasion dont need a Raid array. From there we'll let everyone else makeup their own minds with a small utility and a couple of screenshots i took using my system. I do browse the web,access my email,chat on our forum and use JASC Paintshop Pro7,and my wife uses MS Office. However 90% of time i use my machine for what i built it for, hardcore online FPS gaming; DOOM3, Quake4, Call of Duty2, etc. For going on 4 years now i've run different sizes and arrays of RAID including 0,1,0+1,5 on different PC's. Unlike poor Elton Wilmot I've never had a HD failure,(i use only Seagate ATA or SATA HDD.)or any of the other problems he stated, perhaps because its installed correctly (by me) or the fact the entire operating system is stable even though its overclocked as all the systems i put together are.(Overclocking is truly something most people shouldnt do!)I believe Elton had a whole lot more wrong then a RAID setup! There is WITHOUT QUESTION a noticable performance increase when utilizing a hardware configured Raid array over a single HDD. On July 7, 2005 you stated your thoughts along with afew exceptionally guestionable examples of under researched articles and the statement by AnandTech,(a pinnacle of knowledge)to draw a conclusion thats far from correct. Over a year before your statement a small utility was out that anyone could use to test HD performance and compare ATA or SATA or SCSI in Raid and observe how other operating systems utilize HDD's IO's, heres a link. www.simplisoftware.com/Public/index.php?request=HdTach. Here is 2 different screenshots 1 with and 1 without a comparison to a single drive of the exact same type and number of my HD's. Notice the sequential read speeds,the average read speeds and the most telling difference the burst speed. http://www.putfile.com/pic.php?img=4354255 ,and http://www.putfile.com/pic.php?img=4354256. Also both of my drives are Generation 1 SATA @ 1.5GB/sec the new Gen.2 at 3GB/s blow these away at over 100GB/s sequential read speeds. There are far more articles and benchmarks available online for further documentation of the speed and performance of using a Raid setup then can be listed here....AMD FX-60 @2.92, ASUS A8N32-SLI Deluxe NForce4, 2x1024 Corsair 3200 XMS Pro, 2x80Gb Seagate SATA in Raid0, 2xNvidia 7950GX2's Quad-SLI, 21" Sony Trinitron, WC 101 Watercooling & custom mount 5700rpm Thermaltake fan, Cavalier3 Coolermaster case, BFG Tech 1000 Watt PSU, LiteON CDRW SOHR-5239V, LiteON DVDRW SHW-160P6S, USB LiteON DVDRW.

Steve Lewerer on December 28, 2006 04:00 AM

Well Well.
I have had 2 WD 150 HDs setup via raid 0 for a whopping 300 gig. with My OS on it.

Flawless...until I ran my own vent server and GuildWars at the same time...Numerous bluescreen reboots. and the last one I just had, states Unable to detect one of my raid 0 drives.

I have a msi 875p FIS2R neo

now I am concerned that ll my music, photos, everything has been lost.
and I do have an external 500 esata that I forgot to back my music up to..anyhope to recover data through bios o anything? The drive works, no sound, it just shows up as being "not detected"

mangostain@yahoo.com

oh crap on January 27, 2007 02:05 PM

The cheapest setup to offer some data security is to partition a single drive one partition for windows and other disposable apps that can be reinstalled. The second partition would contain all your downloads photos etc so if things do go belly up it will most likely only damage the windows partition.
Better again is two seperate drives and it is not hard to write a simple batch file to backup all your my documents folder to the seperate partition or drive using robocopy that with the right parameters can quickly backup only new and changed files.

Damien on February 5, 2007 05:24 PM

I'm seriously thinking about either running raid 0 or getting the WD Raptor drives... but all this information and conflicting opinions just confuse me...

I'm in process of building an extreme set up,

EVGA 122-CK-NF68-AR 680i Sli Chipset Motherboard
Intel Quad Core QX6700 CPU or X6800???
Corsair Dominator 2X2048-6400C3DF DDR2-800 RAM
EVGA nVidia 8800GTX KO w/ACS3 GPU

As far as i can tell that's pretty much the most extreme set up; non-overclocked, i can get on paper. Now with such a system i'm seriously worried about bottlenecks, wouldn't it suck once i spent all that hard earned cash and found my HDD's let me down something terrible!!!

I'm going with SATA-300 at least, but with all this contradicting views and info, i'm well confused, there doesn't seem to be any real tests performed on obvious things, file transfers, boot ups, game loading speeds, program loading speeds...etc, all just program running performances, but surely thats when your RAM's more important than your HDD??? I want things to load quick, save quick, write quick, read quick, and transfer quick!

Also I've seen the WD-RE enterprise drives which are gurranteed to an average of 1-Million hours, surely a couple of these drives in RAID 0, and a strict back up routine would prove more benificial than a WD Raptor drive or even just plain old SATA drive.

I can get 2 x 250GB WD RE drives for £50 cheaper than one WD 150GB Raptor, and i've read horror stories about Raptor failing big in the long term.

Also i've never had a drive failure ever, after extensive 5 years of usage... constantly reading/writing large 2-8GB files! When people's RAID 0 drives crash is it because of their maintenance routines? Poor choice of Drives or other equipment? Operating conditions?... or is it just because they run RAID 0???

Any suggestions???

Aidan on February 5, 2007 07:45 PM

Oh yeah i forgot to mention, i was cotimplating running more than just two drives in RAID 0, 3 or 4 160GB drives, would this make a huge difference to all those benchmarks???

Will it kill system performance?

Aidan on February 5, 2007 07:52 PM

I love these threads.. So many folks out there stateing the facts based on some goof ball benchmarks.. Benchmarks are not 100% and only cover the particular circumstances set by the tester. Benchmarks are notoriously accurate but at the same time not real life benchmarks. You get a specific breakdown of a specific point in time based on a specific configuration... Please... if you want to spout off about how RAID 0 sucks.. then try running it for over 5 years. I have had one failure due to a bad drive.. and yes.. I know if it crashes I loose it all... but I still run it.. Why? Number one.. I like living on the edge and I like speed. So qoute your retarded benchmarks on a handful of situations and flame me now... because like everything else in life experience says it all. Benchmarks are for people that one quick results and instant ammo for threads to make the world think they are the next prodigy. Take cars for instance.. Ok we test the new prototype in a controlled environment and get great results.. great handling..speed etc. Ok.. now road test.. uh ohhh.. not so great specs.. The so called controlled environment is no longer controlled.

Granted.. IDE RAID o and early RAID controllers at times I think (based on my experience with all kinds of controllers and setups)were poor performers and yes I think gave bad results regardless of goofy benchmarks wether sporting good performance or poor.

However next gen controllers.. better drives... better boards etc have since really streamlined and elimenated bottlenecks. 90 percent of the computer users don't even know what a bottle neck is other than a bud light etc. I have always seen the Hard drive as the major bottle neck for yearssssss! Thats why I started messing with RAID.

The bottom line.. I'm not a tech sitting at a bench running some benchmark software.. I have, but I prefer real world scenario's. To drive a simulator is not the same as being on the road so to speak.
I live and breath computers and my Biz is 10 years old. I mainly work for companies.. I have 25+ years in gaming and yes.. my machine is tweaked to the max and I upgrade 3 or 4 times a year.. All I want is performance. Crash if you will... I can reload. fragmented.. junked up with files you name it.. my pc still loads the same. Fast.. We have LAN parties etc.. and me being the only RAID 0 guy.. well I am the first on the arena almost everytime and being the first to load and get on the map is crucial at times.

So flame me now.. I have no benchmark proof.. all I have is 10+ years of experience and knowledge of hardware on a grand scale. Not to mention many other things. RAID 0 works.. benchmarks or not.. Its not 100% but what is? Sometimes things work better than other times.. some configurations are worse.. some better. Based on at least 5 years running RAID 0 yes.. it works.. and yes there is a noticeable difference.. sometimes more often in some circumstances than others... and sometimes not. if your an enthusiast.. do it. If your a regular joe.. stay away.

What everyone needs to be fired up about is overclocking. You want instabilty? Do some overclocking.. see how many times you crash before you fly.. and then let the temps or humidity change a little and crash again.. Pushing your system to the max for such little gain and such huge amounts of wear on your components is not to bright.. Overclocking is purely bragging rights.. go ahead and spout off some killer overclocks.. none will impress me because your prematurely killing your machine.

Sorry for the lecture.. but I'm sick of the clowns sitting at their pc quoteing sites and benchmarks.. Most probably never even ran RAID 0 or if they did.. they were biast from the get go and stopped when the pc didn't lift off and fly away.

Forget benchmarks.. be the benchmark.

LordMax on February 19, 2007 09:56 PM

I am also looking at setting up a raid 0 config. The other plus for raid is when you want mega storage capacity in a single volume. It seems that the price for storage (at this point) starts gaining at around the 300gig+ mark. So if you need a huge volume (500 gigs+), the price difference between one massive drive and 2 cheaper ones is about equal, with the raid giving better (if marginal) performance.

So many comparisons also list only 2 drives in a raid 0 config. I am looking at putting 4 250 gig drives into a raid 0. I would think that the performance boost is highly dependant on the quantity of drives splitting the work.

I am probably semi unique in that I play with huge applications, then trash them a month later (current project involves VMware with 8 virtual machines, 3 virtual routers, and a virtual load balancer).

And from a backup point of view, my critical items (taxes and the like) are on a separate machine that gets backup up after every use.

MikeK on February 23, 2007 02:42 PM

Lower extraction times is one of the major benefits I have found with raid 0.

Also since many games accsess the hard drive like crazy on the fly, you can't just quote map loading times. It will only be a perceived slowdown in certain situations, but they do happen in the midst of actual gameplay.

Elder Scrolls is a fine example of a less then optimized game that really taxes hdd's.

Nothing of real value is ever on this raid, I use various ide dumps for that.

Here are some real world mb/s numbers:

//begin quote

Anyways, preliminary benchmarks using PC Wizard 2006:

Before: (Western Digital 160GB SATA2 w/NCQ)
Sequential write: 41.6 MB/sec
Sequential read: 51.13 MB/sec
Buffered write: 133.59 MB/sec
Buffered read: 165.28 MB/sec
Random read: 33 MB/sec

After: (Western Digital & Hitachi 160GB SATA2 w/NCQ drives, RAID-0)
Sequential write: 58.99 MB/sec
Sequential read: 91.92 MB/sec
Buffered write: 172.51 MB/sec
Buffered read: 193.25 MB/sec
Random read: 32 MB/sec


So according to the benchmarking software, I basically improved by a nice margin across the board:

Sequential write: +41.8%
Sequential read: +79.8%
Buffered write: +29.1%
Buffered read: +16.9%
Random read: -4%
Avg: +32.7%

//end quote

reference:
<a href="http://forum.oscr.arizona.edu/showpost.php?s=532b3fdc1cad0246bf89a5dd807cc000&p=11508&postcount=13">http://forum.oscr.arizona.edu/showpost.php?s=532b3fdc1cad0246bf89a5dd807cc000&p=11508&postcount=13</a>;

Mike on February 27, 2007 03:08 PM

I too am considering a RAID 0 setup with two WD SATA RE drives. I have an external drive and will do an OS image as well as regular backups of my "critical" data. I understand the risk but think it is worth it. I've never personally had a complete hard drive failure but I've seen it several times with friends and family. To the guy who is wanting to do a RAID 0 array with 4 drives... that is a risk I don't want to take.

In defense of Jeff, the benchmarks you see for RAID 0 will not always translate to a realized gain in performance. Even though the throughput is higher, your computer is not constantly hammering on the HD (or better not be!) Unless you have a small amount of memory, often the real work will be going on in RAM. For normal day-to-day work your HD will be accessing relatively small amounts of data to process in memory.

I can, however, see the benefits during bootup, defrag, large installs, video processing, high-end gaming, etc.

I think it is a personal decision based on your type of PC usage, your technical ability, and your tolerance for dealing with the risk, setup issues, etc.

Does anyone have any experience with setting up your windows swap file on a seaparate drive from your RAID array? Is there any benefit (or detriment)?

MarkG on April 16, 2007 08:17 PM

I am running windows vista on a raid 0 setup everything runs sweet no probs what so ever and just to let you know My windows experiance score for the hdd went from 5.4 to 5.9 which for me is quite a good improvement and you can see the differance in load times YEAH!!1

Creative on May 27, 2007 11:38 AM

Big Deal; so you are first in the arena. How much sooner? A seC?
This means nothing, other then you spended lots of cash, to gain a few secs.

Hardocp (or was it tomshardwaRE?) have made a nice review, tested loading times for games and overall RAID did NOT help much, if any.

Also, most RAID-wannabee's run it with the onboard crap. RAID can be good, but only when you know WHY you should have it, know what you need for it and know how to use it. Most of the RAID lovers don't come futher then point 3. Lots of them say; for gaming. Now that's a joke, gaming performanve gains are neglectible.

What do you get is much harder configuration to control, less flexible and less reliable. Good luck!


Cyron on June 17, 2007 12:40 AM

My experiences with RAID 0 are very much what you would expect - great "burst" performance for media and virtualization apps (e.g. MS Virtual PC), very little or no difference in gaming (on COD2, Oblivion, Ghost Recon, etc.), but surprisingly boot up seemed a bit quicker with RAID.

My PC is an E6600 w/ 2GB ram, 1 37GB Raptor, 2x Seagate 3200.10 320GB disks. I initially configured my PC with RAID 0, but after reading a few articles decided that the risks outweighed the benefits so I reverted to seperate disks.

Unfortunately, this was on the back of a fairly catastrophic HDD failure in my previous rig, where I lost several months worth of precious photos (happy ending though - managed to replace the PCB on the hard disk and recovered my data!). Lesson learned - I now backup RELIGIOUSLY.

Having sorted my backup strategy, and being an incurable tinkerer, I'm now thinking about switching back again :-)

Meep on June 19, 2007 04:09 PM

I've got to tell you, RAID-0 when using two drives does not double your chances of data loss as so many articles state. This one rumor that got started and spread around without any real math put behind it. If the average failure rate of a HDD is 10% in x-number of years (That's one chance out of ten), then doubling your drives from one to two does not double your risk in losing data. Your risk marginally increases from 10% to about 20% is all. You would literally have to set up 5 drives striped to increase the risk of total data loss to 50% the chance over a single drive. So yes the risk increases some, but not an apocolyptic amount as the naysayers would have you believe.

Apache on June 26, 2007 04:30 PM

In defense of Jeff AND LordMax:

I think LordMax is the exactly the kind of guy that jeff IS NOT speaking to when he writes this article.

It's like when you hear "people shouldn't pick their own stocks" when there are many people who don't do it professionally but can do just fine at it. (and vice versa)

People in this "expert" category will balk at the "conventional wisdom" and warnings like Jeff's, and those found in the articles Jeff cites. Those warnings are to prevent the gadget-hungry psuedo-savvy individuals who know just enough to be very dangerous from jumping on the "double the mean failure time of all your critical data for basically no reason"* bandwagon.

*And that's not a criticism of RAID0, just a real-world example of what a "typical user" can expect from making the change. Outside of gaming (and even then it varies) and anything but the most causal of video-editing, the average person who doesn't know what they're doing should stay away!

Lee on June 28, 2007 11:25 AM

@Brian Hampson:

RAID-1 (or any RAID level for that matter) is not backup. RAID will not save you after you rm -rf * in the wrong directory, nor will it help when a file is corrupted as it's written out to disc.

RAID-1 will only protect you from a drive failure, there is no substituite for a good backup stratergy (except perhaps ZFS?)

Matt Courtney on June 29, 2007 04:42 AM

Looking at Apache's sums, I can't work it out.

In my head
10% = 1 in ten
20% = 2 in ten

I have doubled my chances of a failure.

i.e for every 10 I will have twice as many fails.

Going from 10% to 50% would surely be 5 times as many failures.

Robert on July 10, 2007 03:01 AM

RAID0 is a bad idea, that is the only part that is right. with high failure rates and cheap disks, having a REDUNDANT array of disks is a good idea. a very cost effective solution is linux raid10 or raid5. redundancy in raid is of course no substitution for regular backups.

Mirko on July 13, 2007 11:39 AM

hey ive read all your comments and im going to still try raid 0

luke on October 10, 2007 02:40 PM

. It is amazing to hear all this arguement about RAID. RAID, by any means, is not for the inexperienced. RAID level 1 is okay, if you have critical data you must not lose and have to keep a machine running 24/7.
. Getting to RAID 0; yes, you will see an improvement in performance, and yes, you will see better throughput. If you decide on that, please be sure to have a backup drive of some sort to keep your data safe, for if you lose a drive, you lose it all.
. RAID 5 is something I would recommend entirely. Especially if you have to keep a system running 24/7. Even if you do have a hard drive failure, you can shut down and hotswap a drive, rebuild AS your system remains running. This is crucial for servers. With a RAID 5 setup, you still gain performance over a single drive system.
. I don't quite understand the 0 + 1 plan. I guess if you want superfast performance in parallel, I guess it's ok. But now you are just running two RAID 0 arrays with mirroring. I believe I would still find some sort of backup plan for that.
. All RAID level configurations are serious and not for the inexperienced user. Even with a RAID 5 setup, you should still have some sort of backup plan.
That's my two bits, resume arguing.

Shawn on October 14, 2007 11:45 PM

I tried RAID 0 about three years ago when I built my computer. I disagree about the performance: it was fantastic. *So much better* than a single drive. Mostly I was doing graphic design work, but also doing a lot of gaming.

Alas, one drive error later and I was stuck redoing the entire OS. I knew what I was getting into, though, with the RAID and so I was prepared with a backup. When I redid the system, I decided to forgo the RAID. Result? A *substantially* slower system. But nearly two years after the reinstall, I've not had any reliability issues.

If you're looking for speed, even on a desktop system, I'd recommend a RAID. But go with RAID 5 if you can (it's more expensive to do so, of course).

ern on November 13, 2007 03:22 PM

I've been using a Western Digital 1 TB USB drive for the last two months as my primary repository, connected to an HP Pavilion zv6000 with 1 GB RAM, 100 GB HD.

I have it configured for RAID 1

I've been pleased with it so far.

anonymous on November 21, 2007 05:19 PM

RAID1 is GREAT for desktop--both speed and safety--with a price. RAID1 seems to be confused with RAID0. RAID1 is much better for C: than RAID0.

RAID0 is a poor term! It is not redundant, increases failure risk, and not significantly faster for C: drive (random access).

However, RAID choice depends on your use: random/sequential, read/write.

RAID0 - Bad idea for OS, good for data.
========================================
RAID0 (striping) takes two drives and interleaves data between them to result in one larger drive. It can nearly double read/write throughput, but only if files are larger than one stripe (usually 64/128 KB). It is a terrible choice for your OS (C:) drive because most system files are small, Windows in particular. Every icon on your desktop is a small chunk of a file and requires loading. How about every tiny little registry access? Ever seen the thousands of DLL files in your System32 folder? Largely under 64K. Hear how your drive thrashes wildly at boot? Lots of small files. This is called random access. RAID0 is terrible for random access--one drive does the work and the other sits idle.

RAID0 for sequential access (large files read contiguously) is a good choice! The two drives operate in tandem reading one file. Using RAID0 for large photos or videos that need to be accessed quickly is a good choice.

RAID1 - USE THIS FOR WINDOWS!!!
========================================
RAID1 (mirroring) makes a duplicate copy on two identical drives. You now have failure safety, and it is also MUCH FASTER for your OS. Faster than a single drive, faster than RAID0. Faster than RAID5.

Because of mostly reading many small files, RAID1 is nearly twice as fast because each drive can independently read files at the same time. Use 10,000 rpm (or greater) drives to reduce seek times and both drives will be working together like lightning!

RAID1 falls down during writes when both drives must do the same thing. RAID1 does not help with large sequential accesses because the second drive has nothing to do. Don't use RAID1 for video editing. Use RAID1 for your boot drive! RAID1 is a good choice for data, though, if nothing other than to provide redundancy.

RAID 0+1 - Have your cake and eat it too
=========================================
Combines a mirrored set of striped volumes (4 drives). Best of both worlds, but likely that only two drives will be working at a time (either the striped drives in tandem or the mirrored drives independently). If you can afford it....

RAID 5 - Worst choice of all for Windows drive
==============================================
RAID5 is basically RAID0 with redundancy by dispersing data stripes over 3+ drives. Without a hardware XOR unit, it is slow. RAID5 excels and suffers exactly where RAID0 does. It is GREAT for huge, large sequential reads, less great for writes (because the XOR calculations required), and stinks badly for random access. It's a horrible choice for a boot drive!

If you have money to spend on drives, I'd proceed in this order:
1) http://www.codinghorror.com/blog/archives/000800.html - keep the C: drive small and don't fragment other data into your OS files. A huge drive increases seek times.

2) RAID1 on C:, separate drive for data.

3) RAID1 on C:, RAID0 or 1 for data depending on your data and your desire for safety.

4) RAID0+1

5) Separate drive(s) for swapfile.

If you can only afford one large drive, partition it out so C: is only OS and programs, the other volume for data. That keeps the OS files on the outer part of the drive (which is slightly faster) and prevents inter-fragmentation of data.

Putting "Program Files" on a separate volume makes restoration more difficult and really doesn't do that much--programs install many files in the System32 directory. If you restore/reinstall windows you'll still need to reinstall your programs.

DEFRAG often in any case.

Kevin P. Rice on November 27, 2007 10:53 PM

<i>good backup regimen</i>

Out of curiousity, could you describe your backup regimen, software/hardware recommendations, etc?

Thanks.

Lis Riba on December 23, 2007 03:57 PM

Have run many different raid configs on dozens of systems. For the average home user... it makes no real difference to the experience. They would be better off getting Geeks Squad or someone like that to disable all automatic services and uneeded startup programs.
The only true reason for a raid 0 config is the need for one large volume to ease the burden of data managment. I am currently running 4 Seagate SataII 1TB drives. Drive in port 0 is Vista OS..non raid (backup after install of any new program or after upload of vaction vids or picts etc.
The remaining 3TBs (ports 1-3) are used as my movie server for the rest of the house. I got very tired of juggling movie genres from drive to drive to make an entire genre fit in a single folder on a single drive with another large folder maxing the drive space. 3 Drives, 3 TB no worries. As far as backups... I have the original DVDs anyways.

Jack Humble on May 3, 2008 01:06 AM

everything I read tells me something different, including "expert" reviews. soooo annoying.

even my sisoft sandra says a couple raid 0's would be much faster than a single raptor

ecommerce designer on July 9, 2008 08:43 AM







(hear it spoken)


(no HTML)




Content (c) 2008 Jeff Atwood. Logo image used with permission of the author. (c) 1993 Steven C. McConnell. All Rights Reserved.