• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Anandtech Article: RAID0 not worth it

P

paladin0

Gawd
Joined
May 9, 2001
Messages
632
Well I know a lot of you have been saying this for a while now, but this article made it simple and easy to see how little RAID0 helps (if at all). I probably wont do another RAID0 array ever again. Sure my hdtach score is almost 100MB/s but what does that do for me when loading games nothing.

Sorry if this has already been beaten to death I just liked the simplicity of this article. Also for people who didnt know about this aspect of raid0 this is a good read.

Anandtech Article RAID0
 
Nice article. I dropped my Rator RAID-0 array after a couple of months too. Was useless IMO
 
This is a really great article, and this thread should be stickied. It answers like half of the questions posted in this forum.
 
This article really makes me want to get one of those 74 gig Raptors. Basically equal (or better) than Raid in the real-world tests.
 
My next question is has he tested the performance of different RAID controllers, other than just the ICH5R in the article, to see if they all fall into the same performance levels or run the disks differently.
 
Good article. Actually made me change my mind about getting two Raptors in RAID 0.

I think I'll consider RAID 1 instead.
 
KungFu-tse said:
Good article. Actually made me change my mind about getting two Raptors in RAID 0.

I think I'll consider RAID 1 instead.
Click to expand...

Same thing for me.
Good thing I saw this. I am getting ready to purchase my new system.
 
dainthomas said:
This article really makes me want to get one of those 74 gig Raptors. Basically equal (or better) than Raid in the real-world tests.
Click to expand...


Yeah id rather have 1 74 gig than 2 36 gigs but at the time the 74 wasnt out yet. But now when i upgrade ill have more hard drives for my precious army of secondary comps that keeps accumulating.


Also i noticed people complaining about not using a real raid card for the test. Would this matter much? I dont think the results would be much different. What to you guys think?
 
CIWS said:
My next question is has he tested the performance of different RAID controllers, other than just the ICH5R in the article, to see if they all fall into the same performance levels or run the disks differently.
Click to expand...

nope, however as paladin0 mentioned there are quite a few members that have been pointing out the pitfalls of the real world performance, in both PATA & SATA for quite awhile, basically that its not worth the risk for the marginal performance gains in that level of RAID card or onboard controller (dual channel) and that if you really need that type of STR throughput (video editing, or graphics w\ large files paging to the virtual memory), your likely to be getting a quad or better channel card in SCSI eventually as purely transitional performance space (or alot more RAM or both).

I didnt come around to that viewpoint till I read the As the disc spins series at Lost Circuits
at which point I was able to grasp the difference in the latency overhead and the lack of Tagged Command Queing, that is typical of PATA and SATA and the "typical" controllers
and the "typical" read or write size for most desktop machines

cut and paste
-----------------------------------------------------------------------------------
there is added latency in moving the arms reading the data and transfering it over a single drive
so say you have a large file that is fragemented into several locations, its read performance would suffer because it hits that latency for each location

but say you have the same file contiguously, well your penatized once, but then the faster sustained transfer rate comes in and your alot better off

in a nutshell thats the main difference
and it depends on the number of accesses vs the length of the file in overall performance
sequential reads and writes of larger files will be massively improved
but Random reads and writes of small files will be slower than a single drive
-------------------------------------------------------------------------------------

so I dont doubt the findings at all for the "typical" user
regardless of the controller, there just isnt the need, those large file accesses just dont happen that much for most users
 
CIWS said:
My next question is has he tested the performance of different RAID controllers, other than just the ICH5R in the article, to see if they all fall into the same performance levels or run the disks differently.
Click to expand...
Actually, most other SATA RAID controllers are even slower than the ICH5R. This is because most of them use the overcrowded 32-bit, 33MHz PCI bus with its maximum theoretical bandwidth of only 133.3 MB/sec, and most controllers use much more CPU processing power than the ICH5R. In fact, some RAID 0 setups on one of those other controllers are slower in performance than even a single drive on ICH5/ICH5R. (A few SATA controllers use faster and/or wider-bandwidth buses, but those all cost more than most high-end motherboards, and support at least four SATA drives.)
 
dainthomas said:
This article really makes me want to get one of those 74 gig Raptors. Basically equal (or better) than Raid in the real-world tests.
Click to expand...
Actually, according to your sig, your hardware configuration probably had much more use for RAID-0 than Anand configuration in RAID testing. All "simultaneous multiple data streams from multiple sources to multiple destination" hardware had more use for RAID-0 than typical Intel SEQUENTIAL TIME SLICES ONE AT A TIME Symmetric Processing.

Anand article showed the same thing as typical video benchmarking, the hardware benched very well but don't actually do shit for usage without "simultaneous multiple data streams from multiple sources to multiple destination" supports. Adding sounds and everything went down the tube, adding sounds in Dolby Digital and everything died.

Think of it as a comparison between a hub and a switch, a hub will only connect one source to one destination by broadcasting to everywhere, a switch will actually perform simultaneous connections from multiple sources to multiple destinations at once as long as the bandwidth hold up.

NFORCE series, newer HyperTransport and PCI-E hardware will have greater use of RAID-0 than others. They used the same principle nVIDIA used in the 3 years old XBOX and onboard IGP implementations for gainning performance. They just don't perform benches worthed a shit as most users are just plain ignorant.

I even remember the fools pretending to be expert with no use for "dual channel" and "crossbar". Old Intel architecture do SINGLE TASKING very well, for MULTI TASK they sucks -> no simultaneous data streams for simultaneous tasks.
 
CIWS said:
My next question is has he tested the performance of different RAID controllers, other than just the ICH5R in the article, to see if they all fall into the same performance levels or run the disks differently.
Click to expand...

ATA controllers don't do a whole lot of work, the controllers that actually command the disk reside on the underside of the disk itself. The drive's firmware provides the gigantic majority of the intelligence available and have for many years. While this isn't the best anology in the world, the ATA controller either soldered on your board on stuck in a ISA/PCI slot really doesn't do much more than your average consumer-level NIC.
 
I Think this is a very very good article. It explains a lot about RAID systems and what you can expect from it with today's high performance hard drives, But I can tell you (with experience) that a RAID0 system does have a snapper feel, and with hard drive IO being one of the slowest point of a computer system and also one of the most used for data. Every % gain can be a plus. I do agree there are some down sides and the cost is more for this type of setup, but extra cost is almost always more for high performance setups.

The question here is it worth it to YOU? For some yes, it might be a good idea to get that little extra boost, for most prolly not... I for one liked the extra performance when I switch to a RAID0 setup.
 
Maybe my recent experience is the exception or maybe the onboard Nvidia RAID system sucks on the MSI K8NEO Platinum but I just kicked the tires on RAID 0 with 2 drives and called it quites today. The difference was amazing. Using the onboard RAID controller I was getting an average of 80mb with about 20% CPU utilization when testing with HD Tach. But all the same I was experiencing lag when shutting down and starting programs like UT2004. Nothing huge but a few seconds. In fact, when shutting down UT2004 it took a second or two for the windows desktop to reappear as the GUI was reloaded.

Now that I am running straight drives again, HD Tach shows my OS drive is running an average of 48-50mb with 30-35 CPU utilization and yet, everything is moving faster now, at least when starting and stopping programs.

So between the articles above and this experience, I have to say I am believer that it is useless for desktop PCs.
 
Luthorcrow said:
Maybe my recent experience is the exception or maybe the onboard Nvidia RAID system sucks on the MSI K8NEO Platinum but I just kicked the tires on RAID 0 with 2 drives and called it quites today. The difference was amazing. Using the onboard RAID controller I was getting an average of 80mb with about 20% CPU utilization when testing with HD Tach. But all the same I was experiencing lag when shutting down and starting programs like UT2004. Nothing huge but a few seconds. In fact, when shutting down UT2004 it took a second or two for the windows desktop to reappear as the GUI was reloaded.

Now that I am running straight drives again, HD Tach shows my OS drive is running an average of 48-50mb with 30-35 CPU utilization and yet, everything is moving faster now, at least when starting and stopping programs.

So between the articles above and this experience, I have to say I am believer that it is useless for desktop PCs.
Click to expand...
Did you perform bandwidth management for your RAID-0 configuration? For example, a smaller stripe size will drop the data throughput for a proportional latency improvement.

On classical Intel hardware, there is no bandwidth management. You simply use the largest stripe size possible just at the maximum bandwidth limit. Don't go over the bandwidth limit though, it will immediately cause stutters and data corruptions even possible.
 
E4g1e said:
Actually, most other SATA RAID controllers are even slower than the ICH5R. This is because most of them use the overcrowded 32-bit, 33MHz PCI bus with its maximum theoretical bandwidth of only 133.3 MB/sec, and most controllers use much more CPU processing power than the ICH5R. In fact, some RAID 0 setups on one of those other controllers are slower in performance than even a single drive on ICH5/ICH5R. (A few SATA controllers use faster and/or wider-bandwidth buses, but those all cost more than most high-end motherboards, and support at least four SATA drives.)
Click to expand...

uh, most onboard stuff is also on the same PCI bus... just onboard...

anyway, RAID0 is better when you have a nice PCI-X 3 channel SCSI Raid card and like 6 15k lovin drives or something :D
 
Well, that article @ Anandtech on RAID-0 sure was an eye opener.

I've always had dual HDD's for backup purposes (never ran RAID on them though), and with my last system, I figured I'd try RAID. Looks like it wasn't worth it. :)
 
I could be wrong but I am pretty sure the onboard RAID on this board is part of the native chipset and not part of the PCI bus.

I used 32k. Yes, it did stutter. I suppose if I had played with it and got the stripe size right the problem would have disappeared but judging by the recent articles, it sounds like it wouldn't have been worth the time.

Does anyone have an example under desktop operations were RAID 0 would actually be a serious benefit?
 
Luthorcrow said:
Does anyone have an example under desktop operations were RAID 0 would actually be a serious benefit?
Click to expand...

well some do video editing on the desktop
that would be the type of read & write access that would benefit
but youd likely transit to a multi channel array and RAID controller card
sort of depends on the resolution of the video your dealing with and how close to real time editing it is
 
Snugglebear said:
ATA controllers don't do a whole lot of work, the controllers that actually command the disk reside on the underside of the disk itself. .
Click to expand...

This I would have to disagree with slightly. There is some differences in performance between various manufacturers of controllers and their drivers. I know my onboard ICH5R does slightly better than the onboard SATA RAID with VT8237 and I'm sure people like Promise, LSI, and Adaptec would argure there is a difference. Never the less it's always nice to see results from more than one piece of hardware, maybe Anand will get curious too. Not that I expext the overall results to change, but are curious about any possible differences.

I used the day yesterday to revert from RAID 0 to RAID 1 and can't really notice much in the way it "feels". I may run some synthetic benchmarks, but we all know how real world useful those can be ;)
 
It's sad to look at this article. My future was just blown to pieces. I was thinking of having 4x120gig 8mb wds on raid 0 or 5. Hmm damn you, damn you all :)
 
and why would it effect RAID5?
its not exactly known as a performance level
 
Just got my Oem disk today for the Rig I built, so now I can get a legal and stuff. Once it's massively tweaked, registered, and ghosted, I plan to try other variations with my SLI controller and the 4 raptors I put in this system. I have already tried Raid 0+1 with 4k blocks. System built for Divx encoding and it's already impressed me, topping 70 fps encoding rates. Thanks to the many links provided by Ice Czar, I'm slowly learning about raid , and the benefits to video freaks. Gamers may not see a benefit, but I have.

Can't wait to try Striping the whole array in Raid 0 (kind of redundant since striping and Raid 0 mean the same thing). Keep them great links coming Ice Czar, I may someday actually know what I'm doing. Till then learning from my mistakes.
 
LOL its how we all learn ;)
I dont really understand something till Ive busted it a half dozen times
 
Ice Czar said:
and why would it effect RAID5?
its not exactly known as a performance level
Click to expand...

Supposedly raid 5 offers the best of both worlds. It has the reliablity and kind of matches the speed of raid0. Is this correct>?
 
Not to the same extent. RAID5 has good read throughput, but writes are somewhat lacking. Parity calcs take up a significant amount of power, either CPU or through a processor on your controller. Either way it can only work so fast, and does add significant latency. Additionally, since writes are heavily buffered/cached, you can run into big issues with power and machine failures.
 
I'd like to pound another nail into RAID 0's coffin, and address the myth that RAID 0 is well suited for desktop video editing. I do video editing as a side job, and my experience is with NTSC/PAL and DVD resolution video sources. RAID 0 for desktop editing is pointless in my experience (and I have tried it out).

For re-encoding jobs on single CPU systems (converting video to a different format, filtering, or changing video resolutions or de-interlacing), the CPU+memory is the bottleneck and a single hard drive has no problems keeping up.

For simple muxing jobs, where the audio and video tracks are merged after editing, multiple un-RAIDed drives far outperform a single RAID 0 setup by a factor of 2, worst case. The performance gap widens as more tracks are muxed together. This is assuming that the tracks are located on separate drives, of course. Multiple RAID 0 arrays would work well, although a system with 3+ arrays would not classify as a "desktop" anymore. At any rate, muxing hardly takes any time at all compared to the processing, so RAID 0 is a wash there, as well.

NLE tasks, where the video and audio tracks are constructed from the source(s), does not benefit from RAID 0 either. How much time the editing takes depends primarily on the person doing the editing, and somewhat on the software being used. In terms of time spent, the person is the bottleneck rather than the hard drive.

Bottom line: For my AV editing tasks, RAID 0 did not increase my productivity at all.
 
After 12 hrs of extensive Desert combat testing not once did anyone beat me in a server..

so my raid zero is doing EXACTLY what I need........


Maybe if you had almost 4 ghz your hd would bottleneck..heheh
 
DC/BF load times are dependent on two things, the amount of memory you have and how fast your processor is. I've loaded it on a Seagate 15k.2, an IBM 36lzx, and an old 60GXP. Between all of those there was only about 5 seconds worth of difference between all of them, and this is on a P4-2.26GHz with 512MB of single-channel DDR. However, if I drop into another game such as Half-Life or CNC Generals, disk performance is the predominant factor, and nobody, not even those with RAID0, beats the 15k.2. Heck, most can't even beat the 36lzx.
 
Bones said:
I'd like to pound another nail into RAID 0's coffin, and address the myth that RAID 0 is well suited for desktop video editing. I do video editing as a side job, and my experience is with NTSC/PAL and DVD resolution video sources. RAID 0 for desktop editing is pointless in my experience (and I have tried it out).

For re-encoding jobs on single CPU systems (converting video to a different format, filtering, or changing video resolutions or de-interlacing), the CPU+memory is the bottleneck and a single hard drive has no problems keeping up.

For simple muxing jobs, where the audio and video tracks are merged after editing, multiple un-RAIDed drives far outperform a single RAID 0 setup by a factor of 2, worst case. The performance gap widens as more tracks are muxed together. This is assuming that the tracks are located on separate drives, of course. Multiple RAID 0 arrays would work well, although a system with 3+ arrays would not classify as a "desktop" anymore. At any rate, muxing hardly takes any time at all compared to the processing, so RAID 0 is a wash there, as well.

NLE tasks, where the video and audio tracks are constructed from the source(s), does not benefit from RAID 0 either. How much time the editing takes depends primarily on the person doing the editing, and somewhat on the software being used. In terms of time spent, the person is the bottleneck rather than the hard drive.

Bottom line: For my AV editing tasks, RAID 0 did not increase my productivity at all.
Click to expand...
For the most part, a fast single drive will do layer or two of DV just fine. Several layers of DV, or going to HDV, and a 2 drive RAID 0 array will help. You can do 3 layers of compressed HDV using Cineform's codec with 2 run of the mill HDD's in RAID 0.

But if you get into HD editing, you're suddenly looking at over 100MB/s of sustained read/write bandwidth, up to 250MB/s for 1920x1080 @ 30 in 10-bit uncompressed RGB format. We utilize disk arrays of 10 SATA's in RAID 5 for this on a PCI-X controller. Not sure what nam-ng is on about, however.

Anand did a good job dispelling a few myths about RAID 0, however.
 
You must log in or register to reply here.
Back
Top