RAID 0 and performance

6

64bit_is_here

Gawd
Joined
Jul 2, 2003
Messages
773
see this review:

http://www.anandtech.com/storage/showdoc.aspx?i=2101&p=11

is it fully true? i was about to setup another raid0 array on my cousin's pc(2xwd raptors 36gb) but now i think i mite not go that way as it doesnt really offer any performance benefits and hampers reliability.:( He mostly plays games.

Id love to see this proven wrong tho. iv been really happy with my raid setup for over a year. however if this is the case i might just break my raid0 array now.

Please discuss this review and give some links to some other reviews of this type. Share your personal experiences with raid0 setups, and mention whether it has helped performance or not.

thanks,
 
faster over large contiguous files, but multiple small file reads? it;s going to have higher access times because it has to sync up data between the two drives. Unless he does lots of work with large files it's probably not worth it
 
he goes to the opposite extreme of 'multi-user' in this test, in other words, he runs exactly one application at a time and uses that as his benchmark.

If that is how he actually uses his computer, than the results are probably valid for him. It isn't how I use my computer, and hasn't been since microsoft figured out how to really multitask...
 
Multitasking means even more disk I/O of a random nature, further decreasing RAID0 performance. I tried RAID0 and found it to be a marketing scam many years ago, and since haven't felt the urge to touch it. People still cling to their placebos, though.
 
Th3KrawL3R said:
Raid0 does only one thing.....put your data at risk
Click to expand...

It also provides a good chuckle when someone nukes their array for whatever reason. This is a very similar situation to "fire hot," you can tell people over and over that fire is hot, but they'll still insist on touching it to find out.
 
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.
Click to expand...
I agree with the part about not seeing any real performance gains.
I had a RAID 0 setup with 2 36.7 GIG Raptors and went to a single 74 and feel that for my use, its faster this way.
The part that I dont understand, is how can they say that the MTBF is cut in half.?
This doesent seem to make any sense. I would agree that is is more prone to data loss, but to say that the drives will now fail in half the time as opposed to them being not RAIDED just sounds like BS.
Can someone shed some light on this for me ?
 
Can someone shed some light on this for me ?
Click to expand...

The MTBF of the array not the individual drives, which will remain the same. The array integrity is based on the combined failure rate of both drives and the controller rather than a single drive alone. :)
 
leukotriene said:
The MTBF of the array not the individual drives, which will remain the same. The array integrity is based on the combined failure rate of both drives and the controller rather than a single drive alone. :)
Click to expand...

Ok But if both drives have a MTBF of one million hours (hypotheticaly) does that mean that the Array will have a MTBF of 1/2 million Hrs ? And if thats the case, I still dont get it. What I am trying to understand is how does the physical application of the drives affect the MTBF?
 
Because of how the logical data is stored. If either drive fails the array fails, and your data goes away. If you have a 3 or 4 drive RAID0 stripe any one of those failing will cause the array and your data get munched. It's not that each individual drive is somehow more prone to failure, it's that you're now having to take more than one chance of failure per million hypothetical hours.
 
better way to put it would be for 1 drives, there's a 1 in a million chance per hour a drive will fail, 2 drives 2 in a million, 3 drives 3 in a million, it does cut it in half each time, it just slightly increases the chances you might lose a drive in any give hour. so if you had 16 drives all with MTBF of 1 million hours, each hour they all run there is a 1 in 62,500 chance of failure. small, but far bigger than before. 1 in 62,500 works out to one drive failing in 7 years out of 16 drives, not too bad considering you will proabably upgrade LONG before then
 
Just be wary the MTBF metric has issues, though. It's not an honest number in any sense for any drive. Oh yeah, even overclockers has a nice writeup on RAID0 and game load times today. Looks like all us people in the anti-RAID0 closet are coming out and wiping our feet on the believers.
 
i dont care what storagereview said about raid 0, i run raid 0,i see a huge performance increase, but im not going to go around recommending everyone to get 2 drives for raid 0 because i know my usage is different from other peoples.
 
there is NO POINT for enthusiasts to run RAID 0...You see very little if any performance increase and when your array goes down, you cant recover it (and people like me get a good chuckle out of it :p )
 
The only reason I've found to use Raid 0 (and why I'm runing Raid 0 as my 2nd drive) is video editing.

I'm constantly reading, copying & combining large 200MB - 1GB blocks block of video data, and using Raid 0 on a pair of WD 120 GB drive was huge improvement over a single WD 120 GB drive

The Raptor drives where too small (I needed 200GB+) and too expensive for me.

Yes, I could read from 1 drive & write to another for better performance during copies, but when I'm just reading a file it would be slower, and having everything on 1 drive makes it easier.
 
im just like the average computer enthusiast, and read millions of reviews, and go by them,

ive just installed two WD 80gb SATA drives in a RAID 0 config, with a Promise TX4 controller in the shuttle in my sig.

and ive noticed game load times, even if its by just 10 seconds or so, that's something i wanted to improve on. . .
of course i wasnt expecting to load ut2k4 maps in mere seconds,

for those that hasnt ran RAID-0 in there rig yet, and lived with it for a while, you can't say that much other than linking here and there,., . experince is the best education
 
how does it put my data at risk? i usually format and setup my raid stripe every year.

seems that there is no loss in going for a 64k raid stripe for him and ill format and resetup his raid stripe every year. even if there is a 1sec adv. ill go for it.:)
 
Again, because if any one drive in a stripeset dies or gets corrupted your whole dataset is gone. The data on any drive is meaningless without the data on all the drives in the array.
 
I agree that mtbf is pretty much crap and doesn't mean much but, with each drive your chances do go up for losing a drive in the array so you better have some backups just in case.
 
acascianelli said:
i dont care what storagereview said about raid 0, i run raid 0,i see a huge performance increase, but im not going to go around recommending everyone to get 2 drives for raid 0 because i know my usage is different from other peoples.
Click to expand...

I had a RAID 0 setup with 2 36 GIG Raptors, I lost a drive and lost my data..... TWICE!!!!!!.
No big deal, it was only games and crap. BUT I now have a single 74, and I truly feel that this setup is faster. I cant explain it, and dont have any benchmarks to prove it, but maybe in my instance, for my use, It works better.
 
I just tried a Raid 0 setup. My thoughts were that for games it was horrible. It would really take a while to close down games like Everquest. I know it was probaly due to the stripe size. I eraced my array and installed the drives normal. Its really not worth it.....unless of course you running a server, but for a gaming machine, its not worth it.
 
Bleh, well, I don't know what to say really.

I have a single 36Gb Raptor, and just yesterday got a 2nd one and put it into raid-0. All I did was image the partition onto a backup drive, and then bring it back over onto the raptors after the raid was configured. I got into windows, defragmented (it was defragmented on the single raptor also), and then started playing around.

Game loading performance didn't seem to get much faster, but overall my system does feel a lot snappier. Being I just imaged the partition, I’d say things are at a fairly level playing field (not likely human error as to why things were slower before). I just know that XP boots twice as fast, defragmenting is lightning fast, copying files is really fast, and overall my system just feels snappier. In fact, I’d go so far as to say that going from one raptor to two was a bigger performance increase for me than going from my WD400BB (2mb, 7200 40gb) to my single raptor.

I’ve used Raid-0 for a while, but this is my first hardware raid. Don’t know what to make of my results vs. the ones in the article, but I’m happy. If nothing else, it’s just a whole lot better for me to have ~70gigs of space on my C drive than ~34.
 
I was considering a RAID 0 setup but after reading that article on anandtech and this thread it appears I would see little to no performance gain in my situation (games games games and games). :)
 
I think 1x 74GB Raptor may be a better option than 2x 36GB, especially if you are worried about reliability and not doing video editing.
 
Comparing the 74gb raptors to the 36 gig raptors is not really a fair test. The 74gb's are Raptor II, and the 36gb's are Raptor I.

A single 74gb is almost as fast as 2 36gb's in software raid 0, not because raid 0 sucks, just because the drive is faster.

I am not an average user, I use my computer for video editing/compression/conversion, as well as games, and I do make backups of my data, to me the risk of running in raid-0 does not outweigh the benefit.

I just bought a scsi raid card, it should be arriving this week. I will do some before and after benches with sandra, hd tach, and dr divx and post them out here.

ps. I will be running two seagate 15,000rpm 36gb drives on a 29320A-R, in u160 mode (damn that 32-bit pci)
 
leukotriene said:
In what precisely? Surely not game loading times?
Click to expand...

Yeah, RAID 0 doesn't improve load time for games, so it must becompletely useless.

If you're gaming, you probably don't need RAID for anything. If you're using your computer for other stuff, you may find that RAID helps substantially.

.B ekiM
 
mikeblas said:
Yeah, RAID 0 doesn't improve load time for games, so it must becompletely useless.

If you're gaming, you probably don't need RAID for anything. If you're using your computer for other stuff, you may find that RAID helps substantially.

.B ekiM
Click to expand...

What? People use computers for more than gaming? Impossible!! :rolleyes:
 
What other stuff? Raid 0's only proven significant usage benefit is the transfer of large files, for video editing or a Photoshop scratchdisk. Some people report a change in "feel", which is, of course, completely meaningless. Experience is no substitute for good data from controlled experiments. Ask any doctor.

EDIT: I forgot, it also gives the l33t increase in synthetic benchmarks.
 
L1Trauma said:
Raid 0's only proven significant usage benefit is the transfer of large files, for video editing or a Photoshop scratchdisk.
Click to expand...

I'm confused.

Are you saying that the only application where the transfer of large files is important is in video editing or using Photoshop? Or the only application where the transfer of large files was proven to be imrpoved significantly by RAID 0 was in those applications?

I thought I read that video editing was really CPU bound; encoding, decodnig, maybe compressing, and so on, happens slower in the processor than the I/O can happen while getting or writing the data to be processed. How did RAID 0 help make the CPU faster? Was this test measuring something else?

Who wrote this proof, by the way? Do you have a link to it? What definitions did they use for "significant"? And for "large"?

I love HardForum. It makes me question everything, and I learn so much more that way!

.B ekiM
 
Anandtech and Storage Review wrote the articles. Search for RAID on their sites. Storage Review's is still on the main page.

Significant is my word choice. I don't view a few seconds diffence in load times for some games to be significant. Others view it as very significant.

Large files = bigger than normal Windows/gaming/burning/ripping use.
 
L1Trauma said:
Large files = bigger than normal Windows/gaming/burning/ripping use.
Click to expand...

Like, a 4.2 gig DVD image? That's what I'd call "medium". A 700-something megabyte CD image? That's not really big. An individual MP3 file, at 10 megs? That's not really much.

It's this article that you're referring to, then? It doesn't prove anything; it's a set of measurements.

The point the article actually does make is quite sensical, and very hard to disagree with: "Dont assume RAID 0 offers increased performance for all or even most applications... ".

.B ekiM
 
Interesting info. I run RAID 0 on 2 WD 120g SATA drives. I don't really do anything more with my PC other than gaming, email, IM and web browsing. I may drop the array and just use 2 seperate drives instead. What kind of performance increase/decrease could I see if I ran RAID 1 instead versus standalone drives?
 
RAID 1 is a waste for non-server computers

just run them with no raid

I have the same setup as you, but i think I'll keep my raid. I like having 1 big disk and play around with big files a lot of the time.
 
If you are frequently copying DVD images from HD to HD, perhaps. Burning/ripping? Limited by the DVD drive, not the HD's speed.

The only *objective* data that I've seen regarding actual performance of RAID, not Sandra scores, are the Storage Review article and to a lesser extent the anandtech article. The Storage review forums further dissect why RAID has no benefit to a typical desktop user. The counterarguments, far from being objective data, tend to be subjective descriptions of changes in "feel" and the like. One article did show some decreased UT2K4 load time, but not in Far Cry. The expense plus the increased risk simply overwhelm the minimal-to-nil performance gain for a typical desktop PC user. This contrasts to video editing, which is frequently cited as a RAID 0 advantage; controlling for all other factors, the increased STR results in notable increases in speed for this application.

It's your money. Spend it for that 4 seconds in UT2K4 if you want. If you edit videos or work with big Photoshop files, go for it!
 
raid 0 doesn't increase cpu bound encoding speeds. It will though, allow you to work with raw video off the drive faster, as in loading it to ram.
 
L1Trauma said:
If you edit videos or work with big Photoshop files, go for it!
Click to expand...

Are you sure there aren't more applications than just these two which will benefit from Raid 0?

.B ekiM
 
I know it's not directed at me, but for the most part anythign that would use large contiguous files would be ok, as long as there's not alot of random reads. That's where the big performance loss for raid 0 is from my experience.
 
defakto said:
I know it's not directed at me, but for the most part anythign that would use large contiguous files would be ok, as long as there's not alot of random reads. That's where the big performance loss for raid 0 is from my experience.
Click to expand...

Yeah, that's what I would figure. And I'd think that "large" would really be pretty small -- a few times the stripe size, for example.

I'd also think that some random I/O would be helped, if you've got some good numbers. Say a seek is issued on one spindle, and a read is coming from another. While the read happens, the seek is completing. Then, the seek completes and that drive reading, while the first drive is done reading and starting another seek.

Also, what limits Raid 0 to exactly two drives? If there are lots of drives, it seems like you'd get much better performance for random files, since you only need one drive to be reading while others seek to be ahead of a single drive, which can only seek or read at a given time. The StorageReview article seems to have left this configuration out of its "proof".

But what I'm trying to figure out is why L1Trauma thinks that only Photoshop and videos are helped. He said these were proven to be the only applications where Raid 0 is useful:

Raid 0's only proven significant usage benefit is the transfer of large files, for video editing or a Photoshop scratchdisk.
Click to expand...

and that just doesn't make sense to me. Certainly, Raid 0 does't magically make everything better. But I would expect that lots more applications benefit from it than just Photoshop (specifically) and video editing (in general).

.B ekiM
 
You must log in or register to reply here.
Back
Top