Three WD VelociRaptors get setup in RAID 5 array, testing ensues
Western Digital's hasty VelociRaptor already got reviewed by its lonesome, but for those thinking of getting a RAID system into their rig, HotHardware has taken a trio of 'em, setup a RAID 5 array and put the drives through their collaborative paces. The configuration was made possible thanks to an Areca PCIe X8 hardware RAID card, and the results were rather impressive -- to no one's surprise, might we add. Across the entire volume, performance was generally linear save for a few small valleys along the way, burst speed was 598MB/sec and average read speed was 209.4MB/sec, which pretty much blew the doors off of everything that came before it. Number lovers, there's more where this came from in the read link below.

















Reader Comments (Page 1 of 1)
michas_pi @ Jun 29th 2008 2:29AM
Wicked fast.
thedesolate1 @ Jun 29th 2008 5:05AM
was this post hidden or something? Its the first time I see it.
ShadowKain @ Jun 29th 2008 12:15PM
If only I could avoid them AND a decent hardware based RAID card...
Ignatius @ Jun 29th 2008 3:39PM
Avoid? Do you mean afford?
phanbouy @ Jun 29th 2008 12:37PM
cool.. i wanted a new RAID 5 grilltop
Wolfticket @ Jun 29th 2008 1:13PM
I still think that it is only a matter of time before solid state takes over in the high speed storage niche, but this is one impressive last hurrah for rapidly spinning magnetic platters.
Kris @ Jun 29th 2008 2:07PM
Solid State will eventually take over (it has to). The problem is manufacturing costs and the current reliability of Solid State devices.
Currently it's too expensive to manufacturer a solid state device compared to a traditional hard drive. They're also not as reliable right now as many are rated for a certain amount of reads and writes.
It'll get there but I think it'll be another 10 years before we really can start switching.
JMMGoalster @ Jun 29th 2008 2:15PM
@Kris
but we have to remember, HDD as we know them now weren't perfect when they came out either. They weren't nearly as well done as they are now. It will be a slow phase in, but it'll require us (consumers) to buy them and not just say "I'll wait til they reach HDD levels" because then the companies will go out of business for lack of business.
Peter @ Jun 29th 2008 2:21PM
I agree SSD will take over for limited high-speed storage needs, but it's going to be a while before it can beat magnetic storages' price per Gig.
Especially for backups or other data that doesn't need to be accessed on a regular basis, SSD just doesn't make economic sense.
messalineapghar @ Jun 29th 2008 7:55PM
SSD will take over when read/write cycles are not limited as they are now. which is coming quickly. you will see higher MTBF SSD drives this year, late 3rd, early 4th Quarter.
DefPo3t @ Jun 29th 2008 1:17PM
3 VelociRaptors $900
2 GTX 280 FTW $1399
1 690i SLI ULTA $370
1 PCI-E RAID 5 card $290
1 Intel Core 2 Extreme QX9770 Quad Core $1500
1 Cool Master case $800
1 Prometeia Mach II $800
Having the bigest e-penis while being the dumbest douche priceless
bartoron @ Jun 29th 2008 1:47PM
Indeed.
Dustin Frazier @ Jun 29th 2008 2:16PM
Are you that full of hate? Some people don't mind spending a few thousand dollars on their crack/hobby of choice every few years. In fact, most would consider that cheap as far as hobbies go.
Josh @ Jun 29th 2008 7:05PM
The GTX is a poor choice imo, the Radeon 4870 series in crossfire is better, and also takes advantage of better chipsets for overclocking - ie not being tied down to Nforce.
And there's no DDR3 on that list either >.O
Kate @ Jun 29th 2008 1:41PM
"Across the entire volume, performance was generally linear save for a few small valleys along the way, burst speed was 598MB/sec and average read speed was 209.4MB/sec, which pretty much blew the doors off of everything that came before it."
The raptors have found a new way to open doors? Better warn Dr. Grant...
Mr. Fax Sender @ Jun 29th 2008 2:27PM
You, sir, are my hero.
Kate @ Jun 29th 2008 8:08PM
Sir?!
Jeremiah Zachary Hoffman @ Jun 29th 2008 10:07PM
if xkcd has taught me anything it's that an inate fear of velociraptors is a good thing.
Zzephyr @ Jun 30th 2008 2:55AM
"Sir?!"
Oh, you're not in the military, then?
Jaimi @ Jun 29th 2008 2:14PM
Why bother with RAID 5? It's main benefit is redundancy, not speed. Go with RAID 10 if you have money to burn on this kind of stuff - then you get speed and redundancy.
dfn_doe @ Jun 29th 2008 2:48PM
No doubt, I was just explaining this to someone the other day. Although the benchmark numbers that are posted don't show any write or rewrite tests, which is where the failings of raid 5 become most apparent.
I'm sure that most people would be more than satisfied with a 4 disk raid 10 using commodity 7200 or 10k drives.
messalineapghar @ Jun 29th 2008 7:55PM
The benefit of Raid-5 is that drives can be hotswapped on the fly, rebuilt on the fly, AND the more drives in the array, the faster the array gets. and that speed boost is exponential.
Raid-0+1 or raid-10 , drives need to be added in pairs. a more cost effective solution is raid -6 (3drive (or more) raid-5 stripe, plus a raid-1 mirror)
A.C.E.R. @ Jun 29th 2008 9:13PM
*breaks out Wikipedia info to try to look smart*
Jaimi @ Jun 29th 2008 9:21PM
messalineapghar: you are confusing RAID 10 with RAID 0+1. RAID 10 is RAID 1+0. It's a mirror of stripes - You can hot swap it just fine, and you have to lose 2 (both of a pair) before there is any data loss. Rebuilding a single spindle in a RAID 10 is much faster than a RAID 5 rebuild. Plus, reads are faster, and writes are significantly faster. RAID 5 or 6 has no benefit over RAID 10, except for cost -- it's definitely cheaper, but for most, the redundancy and speed matter much more than the cost of an additional set of drives. (and don't forget to pack in a couple of hot spares, regardless of the RAID level you use!!!)
neal @ Jun 29th 2008 2:22PM
hahaha well said kate ^-^
Eric @ Jun 29th 2008 2:56PM
They should have done this in a RAID 1-0, it would have blown these numbers away.
michas_pi @ Jun 29th 2008 4:12PM
"She's gone from suck to blow!"
saq @ Jun 29th 2008 4:51PM
Sigh, I hate people who use HDtune and HDtach to generate what they think are these really incredible numbers. Those programs are so pathetic at estimating disk performance that they are practically meaningless lies.
If you want to do a real benchmark of disk performance you use IOMeter. Its a lot more complicated to use and understand, but ultimately is one of the only ways to generate true disk performance figures.
Matt @ Jun 29th 2008 8:17PM
Ding ding ding!
IOMeter or simple real world benches (boot time off a fresh install, etc.)
Jon @ Jun 29th 2008 5:28PM
i have been running 3 150GB raptors currently in raid 0 since Xmas, Seems pretty fast ..
dfn_doe @ Jun 30th 2008 1:49AM
Until you lose all your contents due to bit rot. Stripping across disks with no mirror and no parity is an incredibly bad idea unless you are using the disks for a task where loosing all your data isn't going to be a problem. For every disk you add to raid 0 array you increase the probability of losing data because the failure of any single component in a raid0 results in a 100% unrecoverable loss. Such That for a given disk with an MTBF of X you will get X/N where N represents the total number of disks in the system. All other metrics of failure scale identically.
Sean @ Jun 30th 2008 2:37AM
Most of us know that RAID0 = less reliable...these drives are primarily marketed to gamers (like me) who want performance over reliability.
I have been running two 150GB WD Raptors in RAID0 since they first came out. No problems so far, but I do backup to an external drive every night.
If they fail, no big deal, it would give me a great excuse to upgrade to the new VelociRaptors. :)
dfn_doe @ Jul 2nd 2008 2:28PM
@sean
This one always puzzles me. I can't see how having an super fast disk array is going to noticeably improve most games as the majority of new games render content on the fly most of the time and don't do much disk loading. And if you are going to compromise the reliability of your data anyways for faster load time it would seem like you could just keep the game data on a ram disk and obliterate any load time performance edge you'd get from the raid0 of 15k drives. The argument just doesn't hold up; unless by "gamer" you mean "benchmark drag racer" which I guess is just as valid as any other reason, but lets be realistic about our reasoning here.... I know that I built a screaming water/TEC cooled OC rig back 6 or 7 years back and it included a raid0 array, but I was solidly more interested in seeing how good I could get my artificial benchmark scores...