RAID0 doesn't offer performance benefits!

[Áedán]

Shock, horror! Anandtech did a bunch of research with Western Digital's high performance SATA drives, and discovered that RAID 0 offers somewhere between a slowdown, and a minor increase in speed when used on a desktop machine.

Generally the increase was between 2-4%, except for Far Cry, which showed a decrease in performance when loading from a RAID 0 array! Unreal Tournament showed no difference between single drive and RAID 0.

So, to sum up, RAID 0 offers advantages for disk benchmark programs, but makes fairly little different to real world programs.

You can read the article here

[SteveI]

Odd... I'd expect a pretty big boost in terms of large file transfers. Interesting.. I wonder if the system is capable of (or optimized for) the throughput potential of 2 raptors.

Then again, what they are benching in particular might be questionable. Edit: yup. I hate these types of articles. Bottom line: when you raid0, your max throughput doubles, that's it. There is no improvement in anything else, from video card performance, memory performance, CPU performance... even disk access times... so overall, going to RAID isn't going to double overall performance of your system... DUH!

[Gizmo]

My own experience indicates that RAID0 does indeed improve performance, although I don't use it because it isn't redundant. However, I've never built a RAID other than RAID1 that didn't have at least three drives. It might very well be that you have to have at least three drives before the performance benefits come to bear. Or it might be something in Intel's RAID implementation.

[SteveI]

Originally Posted by gizmo

.... It might very well be that you have to have at least three drives before the performance benefits come to bear...

Nah. A couple of years back, I ran a raid0 w/ 2 maxtor 7200 w/ 2meg cache. The improvement was quite noticable to the naked eye. The file loading was really excellent. I think anand is fishing for something controversial, and using tests that don't really measure throughput to stir something up... I guess there really isn't much in terms of new HW worth talking about lately.

[Áedán]

All the sites I've seen, including Anand, have generally found that RAID0 doesn't offer more than a 3-5% boost except on linear reads. Given the length of time that the drives spend seeking verses the amount of time they spend reading, that's hardly a surprise. Measuring throughput is all very well, but isn't at all typical of what a desktop does! Storage Review have a good article on how they test along that hits on typical access patterns.

Move to tagged command queueing and a multiuser environment, and the story is very different. RAID0 and TCQ show very good performance, it's just that it doesn't benefit single user machines.

[Gizmo]

Hmmm..........maybe TCQ is the difference then. This has existed in SCSI for a while, and that is all I use.

Your point about seek time vs transfer time is well taken; this is one of the reasons that I tend to favor large (64 KB or better) stripe sizes. Much below that, and the overhead eats you up. I believe that Anand tested with 128 KB stripes.

[Áedán]

TCQ helps greatly in a multi-user environment, but in a single user environment, where there is high locality, it's a bit of a liability. It's a lot of overhead for little gain basically.

[shadowdr]

I think the problem with the article is that it gives no benchmarks that apply to disk access.There could be a problem with the setup, but if there was how could you tell? "Typical Desktop Use" is statement that applies to many benchmarks that are MS based office applications and in general office use.Not that that is a bad thing but it really doesn't apply to a home user that doesn't own any office applications and might use their computer for video or large file transfers.A more useful article is located here .It also is related to typical desktop use,but I believe them to be a little more informed about the testing methods they use.I also believe that their conclusion was more accurate.

According to the scores produced there was little difference between the raptor and the dx9 that was benched. Some, but not enough to justify the expense of a single raptor.I feel you would be hard pressed to convince a raptor user that his disk is less than 10% faster than a 7200 RPM drive.The only place the raptors were noticeably faster were the Ipeak disk tests.

The article actually raised more questions for me than it answered.Why did they only use sp1 with no further updates when a "typical desktop user" would have all available updates installed? What size partition was used in each setup and would it affect the outcome? Was the partition fully defragged before each of the tests were run or at least between each drive change? Was the page-file optimized for the amount of ram or was Windows left in charge? What stripe size was used? Was it a small one for the OS or was the raid array in fact full of holes?The fact is all of the above could have an impact on the outcome.It would only be a matter of how much.The fact That I have these questions makes me think that the author either did this intentionally or is no expert on the subject.

if you were to ask a raid0 user if they thought their array was faster than a single disk, their answer would be; No sir,I know it is.I can see and feel a difference.Raptors may or may not perform differently in raid0 than a 7200 rpm drive.They would surely produce different results with another raid0 solution(PCI card or chip-set).

Albeit an interesting read, I would not make a buying decision on an article so vastly short on important information.I feel the article was written for the average user not the folks here, and in that respect the author is right on.Raid0 would be of little use to the average joe and a single raptor would be a more effective bang for his buck.

[cadaveca]

uh...RAID....RANDOM ARRAY of INEXPENSIVE DISKS...
kinda goes to show that raid is not meant for "new drives"
just a way to make use of old drives.
a combo of raid 1 & 5 will add a little redundancy to your setup..and should a disk fail..you can get back up and running in seconds.

THAT is what RAID offers. as a benefit it's slightly faster too. too many tweakers looking for milliseconds.

[Áedán]

First up, RAID = Redundant Array of Inexpensive Disks. It was inexpensive compared to the alternatives at the time RAID was a new concept. There's nothing random about a RAID array, which really needs all disks to be identical in order to acheive the best performance.

It's definately not 'just a way to make use of old drives'. It's primary use is in the first word - redundant. People who use a RAID array seriously require that redundancy. They can't live with the downtime a dead disk would represent. For some companies, an hour of downtime represents a loss of more than $100,000. If it took three hours to do a restore, that's a huge loss! Mind you, people who use RAID often use other redundant kit too. Having done work in this area in the past, it's not usual to see redundant routers, databases, webservers,firewalls, application servers, switches, load balancers, power feeds, power supplies and internet feeds.

[robbie]

I just installed my raid0 last night. Tonight I'm going to install BF and BFV. I'll report back on the load times. I can almost bet they they are going to be alot lower then what they were with just my single drive.
Rob

[Áedán]

Originally Posted by shadowdr

I feel the article was written for the average user not the folks here, and in that respect the author is right on.Raid0 would be of little use to the average joe and a single raptor would be a more effective bang for his buck.

Unfortunately, the test suite that Davin and Eugine use at Storage Review is no longer available, and hasn't been for a number of years, otherwise I'd have a copy! SR do provide a far superior source of information, with much thought going through the process. SR and Anandtech are pretty much in agreement - RAID 0 does not significantly boost performance. SR's article on TCQ, RAID, SCSI and SATA pretty much comes to the same conclusion.

[cadaveca]

Originally Posted by Áedán

First up, RAID = Redundant Array of Inexpensive Disks. It was inexpensive compared to the alternatives at the time RAID was a new concept. There's nothing random about a RAID array, which really needs all disks to be identical in order to acheive the best performance.

It's definately not 'just a way to make use of old drives'. It's primary use is in the first word - redundant. People who use a RAID array seriously require that redundancy. They can't live with the downtime a dead disk would represent. For some companies, an hour of downtime represents a loss of more than $100,000. If it took three hours to do a restore, that's a huge loss! Mind you, people who use RAID often use other redundant kit too. Having done work in this area in the past, it's not usual to see redundant routers, databases, webservers,firewalls, application servers, switches, load balancers, power feeds, power supplies and internet feeds.

you said what i meant to say but i had just crawled out of bed when i posted.
lol

[weldzilla]

Well Raid0 array has only one beneficial application that I have found. That is digital video editing. Until I hooked up a Raid0 array with my las system it would not let me download digital video directly onto my hdd. Got a message that said my hard dive was too slow. That was a Maxtor 120 gig 7200 rpm with 8megs of cache. So I got another,raid striped them then I could dump the video in at full qualty fast!! As far as games go, no effect that I could see.

WZ