Sign in with
Sign up | Sign in
Your question

Raptor vs. Raid 0 (4 drives)

Last response: in Storage
Share
March 8, 2007 6:01:25 AM

It appears that I can buy four 250GB drives for the price of a 74GB Raptor.

I know the Raptor is better for smaller files because it has lower access times, and that the RAID 0 is better for large files because its faster.

But can a Raptor really beat RAID 0 with FOUR drives? The performance should be double that of RAID 0 with 2 drives, right?

The only problem is that I already have one of the 250GB drives, but its an 8MB version. So that means the 4 drives will each be 8MB aswell. Will this matter a lot, performance wise? I suppose I could just not use the 250GB drive that i have now.. but then the Raptor will cost less than 4 drives.

Thnx!

More about : raptor raid drives

March 8, 2007 5:52:06 PM

without going into a lengthy discussion thats liable to cause conflicts between posters... you should refer to this thread thats currently going on, about the exact same question (at least if youre intending to game anyhow)

http://forumz.tomshardware.com/hardware/modules.php?nam...

as far as non gaming matters are concerned... if youre strictly dealing with large sized file transfers, or constant (nonstop) hdd accessing... i believe 4*250GB 8MB hdds in raid 0 (128k/largest stripe) will offer better performance, than a single 74GB 16MB raptor will
March 9, 2007 3:52:56 AM

4 drives in raid0 is a ballsy move. You have a significantly greater chance of HD failure.
Related resources
March 9, 2007 4:43:02 AM

its realistically the same chance of failure as having only a single hdd, TBH

in the case of raid, or no raid... you should always have multiple backups, that are updated regularly, so even when a hdd does fail, you dont end up losing vital data
March 9, 2007 4:48:16 AM

I don't know where you are shopping, but I've never seen a 250GB hard drive for $40, and the 74GB Raptor is 4x$40 = $160 on Newegg. You can even get some money back on the Raptor from a rebate.

Either you Raptor price is inflated, or you have one amazing deal on the 250GB drives. The the later is the case, please share with us where you found the deal :!:

Unless of course, you are in a country other than the US and didn't mention it.
March 9, 2007 5:39:29 AM

I've been using these 4 Hitachi 80gb raid 0 for almost a year now. The only failures I've had came from loose cables. It's a gaming rig so no big loss. I've tried various stripe sizes. A 4k stripe gets close to 190mb/sec avg. read. A 128k stripe is about 160'ish avg. read.

So yeah it's a nice little setup; and quite fast too. I couldn't think of going back to a single hard drive even if it was a Raptor. Sure, my seek times aren't as good and cpu usage is slightly up but I don't care. I love it.

Here's a screenie.

http://www.knightsguildbattalion.com/kgb/datas/users/12...
March 9, 2007 7:03:36 AM

Let's debunk some RAID0 myths: RAID0 is good for random I/O too, such as writing small files and doing very much non-sequential I/O access. Following benches are single disk versus 4 disks in 100% Software RAID0 config:

[code:1:00c11e27f3]Single drive (ad8)
concurrency Performance in I/O's per sec. average
1 106 106 107 106
4 106 106 106 106
16 116 116 116 116
32 127 125 126 126
128 151 151 150 150
256 156 156 157 156

gstripe 4xad - 128KB stripe - FM off
concurrency Performance in I/O's per sec. average
1 173 173 173 173
4 270 270 270 270
16 338 338 338 338
32 370 370 370 370
128 444 434 434 437
256 465 465 465 465[/code:1:00c11e27f3]
Conclusion: performance increase ranging from 63% to 300%.

Sequential Transfer Rates:

[code:1:00c11e27f3]Single Drive (ad18)
-------Sequential Output-------- ---Sequential Input-- --Random--
-Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
4096 59391 50.7 59105 16.9 23664 7.0 62471 65.2 63045 10.0 103.7 0.6

gstripe (4 disks on nVidia controller [Embedded], 128KB stripesize)
-------Sequential Output-------- ---Sequential Input-- --Random--
-Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
4096 104288 90.4 237690 74.4 71008 22.0 87837 91.9 250858 44.6 114.8 0.7[/code:1:00c11e27f3]
Sequential read singledisk: 63M
Sequential read 4-disk RAID0: 251M
RAID0 efficiency: 99,6%

Sequential write singledisk: 59M
Sequential write 4-disk RAID0: 237M
RAID0 efficiency: 100,4%

RAID0 scales linear in STR benchmarks. Beware of crappy controllers who use a PCI-bus though.
March 9, 2007 7:59:42 AM

>its realistically the same chance of failure as having only a single hdd, TBH


Oh, Lord, not this argument again. It's *four* times the failure of having a single hdd, not the same chance of failure.

You don't understand basic statistics!


http://techreport.com/reviews/2001q2/realraid/index.x?p...

Explains it better than I can.
March 9, 2007 8:03:37 AM

>RAID0 is Good for random I/Os too

No, it's not. Please can you provide more information on your benchmarks, as you are merely posting figures without any proof behind them.

What machine were these using. What controller. What O/S. How is it duplicable. Etc.
March 9, 2007 8:45:17 AM

Quote:
>its realistically the same chance of failure as having only a single hdd, TBH

Oh, Lord, not this argument again. It's *four* times the failure of having a single hdd, not the same chance of failure.

You don't understand basic statistics!

http://techreport.com/reviews/2001q2/realraid/index.x?p...

Explains it better than I can.


I lol'd. :lol: 

If you toss a coin, do you have the same chance of getting a head as you do tossing 4 coins?
March 9, 2007 8:48:17 AM

In all practicality you're better off with a single raptor or at the most 2 raps in raid. The reason I say this is because 4 drives in raid 0 are hot, heavy, noisy, and messy. Take it from me I know.

Don't get me wrong I like it. But I'd trade them for a pair of raptors just for simplicity's sake.

But just to let you know and I kid you not I load BF2142 maps in under 10 seconds. Crunch that.
March 9, 2007 9:31:15 AM

What would be the sort of things that the Raptor would be better than the Quad-RAID 0 in? Im mainly interested in shortening boot and load times.

Would the 8MB seriously constrain the performance or would it be ok since it would then become 32MB?

Also, three 250GB drives cost as much as a 74GB raptor where I live, in Israel. I got a friend who may be able to get me a 150GB from the US, for the same price as the 74GB here though. So in all, its a 150GB raptor vs. another three 250GB drives in RAID 0.

thnx for ur help!
March 9, 2007 1:10:10 PM

Quote:
>its realistically the same chance of failure as having only a single hdd, TBH


Oh, Lord, not this argument again. It's *four* times the failure of having a single hdd, not the same chance of failure.

You don't understand basic statistics!


http://techreport.com/reviews/2001q2/realraid/index.x?p...

Explains it better than I can.


yes, i know that... with 4 drives in raid 0, the 'theoretical' loss rate is 300% greater than a single hdd... but in real life, youre 'never' going to reliably lose even one of them before the other (though hdds will fail, so thats not in question)... so you have theoretical, and you have real world... if you maintain the hdds well even, just like with a single hdd... the 'chance' of failure, is even less...

and, reliability becomes even less important, if you are not on a critical data system... and you maintain your data through regular backups

but, im not at all recommending raid 0 either... i know where it benefits... and in a lot of typical deskstop situations, it doesnt [provide the improvement thats expected]

just like raid 0s 'theoretical' performance boost is 100% greater for 2 hdds than a single hdd... when real world performance, for certain situations even, is only a *max* of ~15%, nowhere near the 100% theoretical (windows boot times however are an exception to that, due to the way its designed (dont know the specifics of it myself), so it inherantly receives a large boost from being in raid 0)


@Track

but regarding my above posts... i was just exhausted and tired of debating where raid 0 benefits, and how much it does (i went through that whole thing months ago on here, in another thread that was certainly more heated than the above one)... so if youre at all interested in gaming, you should read the above thread i linked to

or this thread (from months ago):
http://forumz.tomshardware.com/hardware/modules.php?nam...
March 9, 2007 4:14:25 PM

Quote:

but, im not at all recommending raid 0 either... i know where it benefits... and in a lot of typical deskstop situations, it doesnt [provide the improvement thats expected]

just like raid 0s 'theoretical' performance boost is 100% greater for 2 hdds than a single hdd... when real world performance, for certain situations even, is only a *max* of ~15%, nowhere near the 100% theoretical

If this is the case then some things are wrong:
- too low stripesize
- filesystem misaligned with stripe blocks

in both cases a single I/O request will cover 2 or more disks, resulting in lower performance since no parallelization is possible anymore. If RAID0 is done correctly, it will speedup random I/O quire nicely, also in desktop environment. Please see my posts above for more detailed info.
March 9, 2007 4:22:06 PM

Quote:

but, im not at all recommending raid 0 either... i know where it benefits... and in a lot of typical deskstop situations, it doesnt [provide the improvement thats expected]

just like raid 0s 'theoretical' performance boost is 100% greater for 2 hdds than a single hdd... when real world performance, for certain situations even, is only a *max* of ~15%, nowhere near the 100% theoretical

If this is the case then some things are wrong:
- too low stripesize
- filesystem misaligned with stripe blocks

in both cases a single I/O request will cover 2 or more disks, resulting in lower performance since no parallelization is possible anymore. If RAID0 is done correctly, it will speedup random I/O quire nicely, also in desktop environment. Please see my posts above for more detailed info.

i dont doubt that really... but for 'most' users, how many of them are going to have constant nonstop random I/O's?... to make much use of that at all?... if youre running a server, sure... that would make sense to help... but the average user isnt running a server though, that needs that kind of nonstop random request performance (or multiple background applications that are constantly accessing the hdds)... the STR % increase is what i am referring to though, due to primarily large file transfers, you will never see that kind of % increase in real world STR performance (not benchmarks).
March 9, 2007 4:32:22 PM

raptor. is. safer. which. matters. more.
March 9, 2007 4:34:11 PM

Quote:
the STR % increase is what i am referring to though, due to primarily large file transfers, you will never see that kind of % increase in real world STR performance (not benchmarks).


Sorry, choir, but that simply isn't true. The performance you can take advantage of in the "real world" is directly dependent on what you're doing.

At work, I have a DVD authoring workstation that is equipped with two separate RAID-0s. When we author a DVD, the build operation copies around 4.5GB of files from RAID-0 #1 to RAID-0 #2, multiplexing them into the .VOB files needed for the DVD.

This operation is almost exactly twice as fast as it used to be when doing a DVD build from single drive to single drive. And the reason for that is very simple - the operation is virtually entirely dependent on sequential transfer rate.

You cannot generalize that "real world" performance is always lower than theoretical performance just because "that the way it is". A properly designed system (not just RAID or computer, but any physical system) can always reach near it's theoretical performance if all factors are taken into consideration. It's a matter of application engineering, not just theory.
March 9, 2007 4:35:32 PM

Quote:

but, im not at all recommending raid 0 either... i know where it benefits... and in a lot of typical deskstop situations, it doesnt [provide the improvement thats expected]

just like raid 0s 'theoretical' performance boost is 100% greater for 2 hdds than a single hdd... when real world performance, for certain situations even, is only a *max* of ~15%, nowhere near the 100% theoretical

If this is the case then some things are wrong:
- too low stripesize
- filesystem misaligned with stripe blocks

in both cases a single I/O request will cover 2 or more disks, resulting in lower performance since no parallelization is possible anymore. If RAID0 is done correctly, it will speedup random I/O quire nicely, also in desktop environment. Please see my posts above for more detailed info.

i dont doubt that really... but for 'most' users, how many of them are going to have constant nonstop random I/O's?... to make much use of that at all?... if youre running a server, sure... that would make sense to help... but the average user isnt running a server though, that needs that kind of nonstop random request performance... the STR % increase is what i am referring to though, due to primarily large file transfers, you will never see that kind of % increase in real world STR performance (not benchmarks).
With STR, a RAID0 setup always increases linear unless you have an extremely crappy RAID0 subsystem or are doing things wrong. Improving STR is not difficult but for 'real life' usage it is not very important either.

More important are the access patterns Windows, WinRAR, Photoshop and various games give you. This access pattern is pretty much non-sequential. A reasonably good way to test this is using random I/O benchmarks; to see how good a RAID subsystem is with less 'dumb' I/O patterns and can distinguish itself from crappy controllers.

The best way to benchmark is either using real applications (such as game loading times) or I/O traces like Intel Ipeak. The problem is that proper benchmarking requires some good knowledge about storage and i see too many unknowledgeable persons/reviewers educating the public about RAID with flawed benchmark setups and wrongly drawn conclusions. Benchmarking on Windows is extra problematic since the old NTFS filesystem has done nothing to prevent misalignment. So the reviewer gets crappy results and concludes RAID0 is to be blamed -- which is a wrong conclusion, it was his setup which is to be blamed.
March 9, 2007 4:37:37 PM

Quote:
raptor. is. safer. which. matters. more.

It does not. Anyone who cares about his/her data will never let data rest solely on one mechanical harddrive. Then you are gambling with your data anyway.
March 9, 2007 4:41:04 PM

if thats true... i stand corrected :) 

i really should reword what i said then, lol... but the average person really wont see that kind of increase though, either way
March 9, 2007 4:46:47 PM

Quote:
if thats true... i stand corrected :) 

i really should reword what i said then, lol... but the average person really wont see that kind of increase though, either way

It's incredible what i read on forums all over the internet about RAID. So many myths that need to be corrected. The problem for this is that there are very few good reviews about RAID out there. I'm working on a huge article about RAID with many benchmarks done properly as i think of it.

For example, people do not believe me when i tell them my 100% software RAID5 does 400MB/s of write performance, in their heads is implanted that using your CPU for parity calculations is slow and you need a hardware RAID card; while it actually beats Areca ARC-1210 regarding STR scores. Though Areca beats badly in non-sequential benchmarks.

But it's one of the many myths that are out there. People don't say their RAID5 implementation/setup is crappy, no they place all the blame on RAID5 itself. That is wrong.
March 9, 2007 4:53:55 PM

the main issue i was trying to prove/disprove, was the actual benefit of raid 0 for game loading times, that i tested myself as well... i really had gotten tired of repeating the same thing (dont know how many times i did, too many), but, if you want to read the above threads i posted you can, and even add that kind of testing to your review... to find out which types of configurations benefit more, and how much... possibly even why... but, its been nothing but an uphill battle...
March 9, 2007 4:59:05 PM

Quote:
the main issue i was trying to prove/disprove, was the actual benefit of raid 0 for game loading times, that i tested myself as well... i really had gotten tired of repeating the same thing (dont know how many times i did, too many), but, if you want to read the above threads i posted you can, and even add that kind of testing to your review... to find out which types of configurations benefit more, and how much... but, its been nothing but an uphill battle...

In my article i will also cover game loading times. One drawback of Windows is also that it has quite crappy software RAID options - i hope Windows Vista to be better in this regard but somehow i doubt it. :( 

For RAID to give a good performance increase the implementation has to be 'intelligent'. All onboard controllers i've seen are 'dumb'.
March 9, 2007 5:07:48 PM

i do have a promise fasttrak s150 tx4 pci sata raid controller... even with positive reviews on it, it didnt seem much different at all compared to my nforce4 onboard raid even... it mightve just sucked i suppose

but, hopefully after your review, more people will know and understand the actual benefits and performance of different raid situations, configurations, and solutions...i can only hope its thurough :) 
March 9, 2007 6:22:12 PM

I don't know why the anti-raid sentiment though. It's becoming a religion. Anandtech and StorageReview started this all.

I bought a 2 port sata raid card several years ago solely for the fact that BF1942 took so long to load. I put a couple of Seagates in raid and load times dropped noticeably. Not huge, but enough to make me happy. For somebody to tell me that this didn't happen is just crazy talk.

Now with four drives I can load BF2142 maps in under ten seconds. And if somebody wants to tell me that's not true then I say they're nuts.
March 9, 2007 6:24:19 PM

Those Fasttrak controllers are actually fake RAID - it's just an IDE controller with firmware bootstrap support and a BIOS utility to store the metadata; the actual RAID is implemented by drivers. So this falls into the same category as onboard RAID.

RAID can be classified in:
- Software RAID (not bootable)
- Firmware RAID / Driver RAID / Fake RAID
- Hardware Assisted RAID
- (True) Hardware RAID
March 9, 2007 6:51:07 PM

hm... ok
March 9, 2007 7:00:26 PM

And sorry if i am being even more negative.. :wink: but the PCI bus is a terrible bus to do I/O on. Not only does it provide little bandwidth and only half-duplex but the fact that this is shared with all other devices (PS/2, USB, Onboard/PCI Audio, legacy I/O, sometimes onboard LAN) is detrimental to real life performance.

I would opt for either PCI-express (which *IS* full-duplex, dedicated, low-latency and high bandwidth) or a dedicated "embedded" bus like the chipset provides to it's SATA/IDE connectors.
March 9, 2007 7:13:23 PM

Dude :D 

Your asking of the question you asked shows you to be relatively non-tech. Your stated problem "Im mainly interested in shortening boot and load times" can be easily "solved" simply be waiting until the Samsung, et al, Flash+Disc composite HD units are released. The Flash mem permits instant (relative to HD accesses) loading the OS-Desktop at boot.
March 9, 2007 7:18:27 PM

its okay honestly... i just reasoned before that the pci bus was limiting it, since it maxes out at 120MB/s for 2 36GB 8MB raptors... and adding a 3rd and 4th didnt make it any faster... i dont think overclocking the pci bus is a good idea... but, would a 33MHz pci card be compatible with a 66MHz pci slot?... cuz it would be nice to get the other 130MB/s out of it... instead of just being a waste of $120, lol

but... if its not compatible (and only fake raid at that), it does definetly seem a pci-e true raid controller is worth considering, especially if there are real benefits to it... theyre probably really expensive though
March 9, 2007 7:58:09 PM

Quote:
its okay honestly... i just reasoned before that the pci bus was limiting it, since it maxes out at 120MB/s for 2 36GB 8MB raptors

But then you are talking about STR. PCI also limites random I/O, with a total bandwidth of just 14,9MB/s. Bandwidth is not everything, there is also: latency. And when a bus is shared the latency may be much much higher.
Quote:
i dont think overclocking the pci bus is a good idea... but, would a 33MHz pci card be compatible with a 66MHz pci slot?

... no :wink:
Quote:
but... if its not compatible (and only fake raid at that), it does definetly seem a pci-e true raid controller is worth considering, especially if there are real benefits to it... theyre probably really expensive though

The onboard SATA/PATA connectors are on their own embedded bus, and provide both reasonably low-latency and good-enough bandwidth. But, if you can afford a real controller like Areca ARC-1210 then you will not be disappointed by it's performance!
March 9, 2007 8:12:37 PM

Quote:
its okay honestly... i just reasoned before that the pci bus was limiting it, since it maxes out at 120MB/s for 2 36GB 8MB raptors

But then you are talking about STR. PCI also limites random I/O, with a total bandwidth of just 14,9MB/s. Bandwidth is not everything, there is also: latency. And when a bus is shared the latency may be much much higher.

that must be where the 15MB/s comes in then...

Quote:
but... if its not compatible (and only fake raid at that), it does definetly seem a pci-e true raid controller is worth considering, especially if there are real benefits to it... theyre probably really expensive though

The onboard SATA/PATA connectors are on their own embedded bus, and provide both reasonably low-latency and good-enough bandwidth. But, if you can afford a real controller like Areca ARC-1210 then you will not be disappointed by it's performance!

yeah... newegg has the areca controller for $320, was actually expecting it to be more expensive ($500+)... as it is right now though, im broke, so if anything, another controller is going to have to wait, lol :) 

edit: not a good thing to be replying to responses when you havent eaten all day... i meant 15%, not 15MB/s... which has nothing to do with the pci bus, oh well, lol (it was just an equation someone came up with that i dont remember)... as far as 33MHz pci being compatible with 66MHz pci... i just figured that since they were both on a 64bit bus, and that it was a common operating speed that might keep things in sync, that it might work (i read somewhere that the 32bit pci bus is actually 64bits wide, but, beyond that, i wasnt really sure if it would)
March 11, 2007 9:11:19 AM

Quote:
Dude :D 

Your asking of the question you asked shows you to be relatively non-tech. Your stated problem "Im mainly interested in shortening boot and load times" can be easily "solved" simply be waiting until the Samsung, et al, Flash+Disc composite HD units are released. The Flash mem permits instant (relative to HD accesses) loading the OS-Desktop at boot.


Non-tech?
A non-tech person wouldnt even know what RAID is. An enthusiest would know access times and transfer rates, and only a person who has studied HDD's for a long time would know about sequential and non-sequintial read times and I/O requests and the rest. So thats absolutely the stupidest argument ive heard in this thread, also because it putting me down has nothing to do with answering my questions.

Im not too knowledged with HDD's, but im far from an enthusiest.
And all i was asking was what were the things that RAID 0 would be better at and what things the raptor would be better at.
December 8, 2008 3:36:13 PM

choirbass said:
Quote:
>its realistically the same chance of failure as having only a single hdd, TBH


Oh, Lord, not this argument again. It's *four* times the failure of having a single hdd, not the same chance of failure.

You don't understand basic statistics!


http://techreport.com/reviews/2001q2/realraid/index.x?p...

Explains it better than I can.


yes, i know that... with 4 drives in raid 0, the 'theoretical' loss rate is 300% greater than a single hdd... but in real life, youre 'never' going to reliably lose even one of them before the other (though hdds will fail, so thats not in question)... so you have theoretical, and you have real world... if you maintain the hdds well even, just like with a single hdd... the 'chance' of failure, is even less...

and, reliability becomes even less important, if you are not on a critical data system... and you maintain your data through regular backups

but, im not at all recommending raid 0 either... i know where it benefits... and in a lot of typical deskstop situations, it doesnt [provide the improvement thats expected]

just like raid 0s 'theoretical' performance boost is 100% greater for 2 hdds than a single hdd... when real world performance, for certain situations even, is only a *max* of ~15%, nowhere near the 100% theoretical (windows boot times however are an exception to that, due to the way its designed (dont know the specifics of it myself), so it inherantly receives a large boost from being in raid 0)


@Track

but regarding my above posts... i was just exhausted and tired of debating where raid 0 benefits, and how much it does (i went through that whole thing months ago on here, in another thread that was certainly more heated than the above one)... so if youre at all interested in gaming, you should read the above thread i linked to

or this thread (from months ago):
http://forumz.tomshardware.com/hardware/modules.php?nam...



My english is crap, try to understand...

Elegant but wrong.
Mathematic is not aritmetic, thats a big mistake and a product of midìa`s huge ignorance.

There is no such absurd as real world statistic and theorical world statistic. Plz note that statistic is the part of math that allow us to aproach our models to the real world.

With that in mind I can comment and try to correct you. Im sure you mean that the chances of losing a single driver are very vey remote, making the chance of losing four driver still little, that a good argument but what the other guy is trying to say is that if the initial chance is x% then the final chances are 4x% and that is correct.

Its basic math, simple like that. The hard part would be calculating the rate failure based on statistic numbers and results since drive failures may be totally random or due a main external factor.

The failure rate is a function were "up time" is a factor, which explains very well why u are saying the chances are the same since the "up time" of 4 drivers will be the same.It is unlikely that 1 would fail after 1 week and the other after 5 years - its more likely that all four fail approx at the same period (their life time) bcuz they are exactly the same model and plus this period would be long enought so you could disre­gar­d­ the difference.

Like I said thats wrong, the failure rate does take "up time" as a fator but the lowest the "up time" less it will affect the failure rate. So innitially the failure is almost totally random so havig four drivers makes those 4x% scary but not enought since 4x%~0.

In a large server or backbone the rates of hardware failure behave like that, when more hardware are added in array (if 1 fails the system fails).

A good and simple exemple that make it easier to understand is this:
- You have a server that is operating at high temperatures, at these temperature the HD failure rate will raise close to 40% (exemple). A server with 4 HDs wouls fail aprox the same number of times as a server with 1 HD ? its obvious that no, a server with 4 HDs would prob fail every 2 minutes of operation, fail -> replace drivers -> fail again (chances are to high and is easy to imagine). Now the server with 1 HD would be able to operate a lot more, fail -> replace HD -> long time -> fail -> replace -> long time -> fail.

You are saying that x~4x bcuz the number is small, thats right but chances for HD to fail are always small. If a certain driver has 1% of chance to fail in the first year then now u have 4% of chances and thats not "theorical" thats real. Theorically you have more then 4% bcuz a more complex system will have greater chance to fail since u have a larger number of components being feed by the same powersuply, a greater number of variations (current, voltage, temperature, noite, eletromagnetic field, etc...) and a bigger absolute variation - vide the link of the guy who posted for you, thats a great exemple of how it works. We use statistic to mesure all this factors and, with advanced math tools (such as calculus, differential equations, linear algebra, statistics) we come out at end with a function that return the significative value. The HD failure rate is the real significative value, obtained using statistic.

Statistic is not easy, is a complex, boring and long class in college and it is indispensable in engineering and abolustly necessary in Quantum mechanics.


Reguarding the post owner question:

Raptors are drivers with exceptional speed and extremely low seek time.
Obviously RAID 0 with 4 250GB drivers are faster then a Raptor and is highly possible that is faster then 2 raptors. But RAID 0 doesnt change average seek time. Random transfer is also better in a stripping with 4 drivers bcuz the seek time on both cases (raptor or 4x250gb) the average seek time can be disregarded.

The cache is also important for both cases and on a stripping configuration, 8MB cache on each disk is far better then 4x8MB in a single disk even with the single disk operating 4x faster and being 4x larger. Imagine a hard disk with 4x250gb capacity, 4x7200RPM and 4x8MB cache, in this disk the cache is less effective then four 250gb in RAID 0, 7200 RPM each and 8mb cache each - it is hard and would be to technical explain why is this but u can imagine that in raid0 configuration the cache would work like a multiple channel cache. That makes it clear that raptor cache wont outperform your raid 0 configuration even being a lot larger then a single 250gb disk cache.

Going 4x250GB with 8MB cache each is obviously worst then 4x250GB with 32B cache each but the difference is small and doesnt worth if u alredy have 1 250GB with 8MB cache. The difference between this 2 systems will rise a lot if the size of the disks with larger cache is also larger. For example: 4x250GB 32MB cache > 4x250GB 8MB cache but 4x500GB 32MB cache >>> 4x500GB 32MB cache.

Like our friend said, with 4 disks in stripping, the space, weight and heat is sure factors that should be studied before getting it.

I was able to fit a CEB motherboard (Asus L1n64-SLI WSB) with
2 processors + 6x HDs, 2x gtx 260 and 1 i-ram on a Antec 900 but it took me ten times the time I would normaly use to build a PC with enought space in the case. The final result were a case weighting 3 tons and a machine that can be used as heater. In this build there is 6x Hds with 500GB each, 4 in RAID 0 and the 2 others also in stripping - the result is a blazing fast system when comes to loading vista 64 Bits or any heavy application and transfer rates between the 2 partitions (2TB to 1TB) can reach 280Mb/s+ when transfering a single file.

Sry for the bad english.

August 23, 2012 10:28:29 PM

choirbass said:
its realistically the same chance of failure as having only a single hdd, TBH

in the case of raid, or no raid... you should always have multiple backups, that are updated regularly, so even when a hdd does fail, you dont end up losing vital data



This is totally untrue!
suppose that we have a disk with a mtbf (mean time between failure) of 1000000 hours this mean that some will fail within the first hours and some will last forever. If you bet all your pot on two drives, you double the chances to get one that fail within the firts 3 years of use. Now if you double that two dives, you'll double the risk the same way. Really simple mathematics and common sens. You'll note that the risk will grow not like 2X but Xexp2 (x represents the number of drives) and that's why huge data center uses raid with parity to lower the risk witout compromising the performance.
!