first RAID setup... any pointers?

E

Everfalling

Gawd
Joined
Jul 12, 2005
Messages
676
i have an A8N-SLI Asus motherboard, and i just purchased 4 300gig maxtor sata150 hard drives. i wanna put them in RAID 0 with the raid controller built into my mobo. My friend said his raid card had a hard time recognizing any drive over 250, will this be a problem with the controller built into my mobo? also, are there any issues i need to think about, be it power or heat or anything? and besdies the risk of one drive failing and taking the other three with it, are there any flaws in RAID 0
 
You won't gain much performance from raid 0, and it increases your chance of data loss by far more than a factor of 4. The limitation is at 131 GB; there's no reason you'd have a problem with 300 gb as opposed to 250. If you've got SP2 you're basically set.

Power and heat will both be concerns. What power supply are you using, and what ventilation is there?

 
I can't offer any advice for your specific situation (board and drives), but are there any flaws with raid 0? Yes. Besides the risk that you seem somewhat aware of, you're not really going to gain much performance (unless this is for a multi-user server with large sequential tasks). A quick search of recent topics in this forum could have told you that much.

More importantly, why purchase 4 300GB HDD's and then come asking questions afterwards? The types of questions you are asking really demonstrate a lack of though to your plans. I hate to say it, but you really should have done some more research before you paid for any parts.
 
unhappy_mage said:
You won't gain much performance from raid 0, and it increases your chance of data loss by far more than a factor of 4. The limitation is at 131 GB; there's no reason you'd have a problem with 300 gb as opposed to 250. If you've got SP2 you're basically set.

Power and heat will both be concerns. What power supply are you using, and what ventilation is there?

Click to expand...


i'm using an Antec TruePower 2.0 TP2-550

ventilation is one fan in the front that'll be blowing past one of the maxotrs and my WD C: drive and one that is just blowing into the case, and the other three are in a case that fits 3 hard drives into 2 optical drive slots and it has its own fan. then i have two exhaust fans plus the PS itself, and one intake fan on the side window.

ekard: yeah...thanks... RAID 0 is just an idea. if it doesn't seem worth the risk i'll just have 4 seperate drives. i'm just wondering which i wanna do. cause i wanna have one big TB drive, and i'm thinking raid 0 stripping would make defragging a TB a bit faster.
 
Sorry, didn't mean to come off as an ass. I'm glad you're looking for the storage and considering raid just as an add on. That said...

Raid O will not be any faster at defraging. Defraging is the copy and paste of many small bits to and from the same drive. Raid 0 benifits the transfer of large, sequential files to and from other hard drives.

I can see the desire for having one large drive instead of 4, but if that's your only reason and you really do have 1 TB of data to store, you won't want to risk all that on raid 0. I've had one scare with raid 0 and it was more than enough to steer me away from it for good. (Though i would consider it for a scratch disk if i got into photo/video/cad work).
 
you mean that you dont think there's any more performace from having 4 drives spin to access a single file, rather than one drive spin for that same file? my assumption is that 4 drives doing the work of 1 is faster, i'm guessing i'm wrong then?

ok, so if raid 0 is too risky just to have one big TB drive, is there another way to do this? is there another way to make as if all 4 drives are one big drive? or would i be stuck managing 4 seperate ones?
 
I know there is a way, but I forget what it is at this time. I'm sure that someone else will chime in with the correct response.

The reason why 4 disk vs 1 is not always faster is due to seek time vs. transfer rate. Yes, RAID 0 will be faster while transfering data (theoretically) because you are splitting the load. However, if the file is small, then most of the time is actually spent looking for the file (seeking). Now with RAID 0, instead of waiting for one drive to seek to the file, you have to wait for 4 different drives to all find the same file wherever they've stored it (they won't all put it in exactly the same spot). For small files, this extra time actually hurts peformance, for large files it's not as big a deal. Still, for those large files, you are assuming that once all the drives have found the file, they don't have to seek anymore. If the file is fragmented on even one of the drives, they all have to slow down while that one drive seeks. This is why single-user desktop results are usually stronger for a single drive. Most strenuous desktop evironements rely more on the pattern of seek-send small file-seek again-send small file.
 
Theoretically, yes. Is the main purpose of the 1 TB drive just file storage? What is the intended use of the machine?
 
just storage. movies, music, pictures. my WD C drive would be seperate from all this. i want to be able to just organize things by folder, not have to divi them up acorss 4 drives manually.
 
Everfalling said:
you mean that you dont think there's any more performace from having 4 drives spin to access a single file, rather than one drive spin for that same file? my assumption is that 4 drives doing the work of 1 is faster, i'm guessing i'm wrong then?

ok, so if raid 0 is too risky just to have one big TB drive, is there another way to do this? is there another way to make as if all 4 drives are one big drive? or would i be stuck managing 4 seperate ones?
Click to expand...


Ok, fine....

you have 4 seeks to make, 4 calculations to make to "put together" that single file and NOW you can load it at 4x the speed (in theory). Well, for small files (< the cache on your HDD) thats wasted time.

Now if this file was 4GB in size, yeah, you'd have yourself a winning situation.
 
You can do what's called a JBOD (just (a) bunch of disks); this avoids the problem of making every drive seek whenever you request a file. There's still no redundancy; if you lose a disk you're still in trouble. But it'll be less wear and tear on the disks, which will hopefully help increase lifetime. Windows (even XP) can do this in software, and that'll be plenty fast.

 
unhappy_mage said:
You won't gain much performance from raid 0, and it increases your chance of data loss by far more than a factor of 4.
Click to expand...
Assuming no dropout errors, then statistically, your failure rate with 4 drives is slightly less than 4 times the rate of one drive.
 
Where do you get your numbers? The chance of a failure is related to the inverse of the sum of the inverses of the component MTBFs:
Code: 
System MTBF = 1/(1/drive1mtbf + 1/drive2mtbf + 1/drive3mtbf + 1/controllermtbf)
So, to take their example, 4 disks at 500k each, and a controller at 300k:
= 1 / ( 1/500000 + 1/500000 + 1/500000 + 1/500000 + 1/300000)
= 88,235
And 88 thousand hours of MTBF isn't as good as it sounds:
This number is very often misinterpreted and misused. Usually, the "analysis" goes like this: "Gee, a year contains 8,766 hours. That means my 500,000 MTBF drive should last 57 years." ... If the MTBF of a model is 500,000 hours and the service life is five years, this means that a drive of that type is supposed to last for five years, and that of a large group of drives operating within this timeframe, on average they will accumulate 500,000 of total run time (amongst all the drives) before the first failure of any drive.
Click to expand...
So if you have 4 drives and a controller, between the 5 of them they'll likely accumulate 88k hours before one fails. That's only 17647 hours, or about two years.

And if you don't believe that, what about extra load on your power supply leading to its early demise? Or a cable coming unplugged and the filesystem trying to "repair" itself and screwing all your data up?

 
unhappy_mage said:
Where do you get your numbers? The chance of a failure is related to the inverse of the sum of the inverses of the component MTBFs:
Click to expand...
The math on that site is incorrect...they used a value derived from dependent events not independent events. The actual calculation for two drives "a" and "b" is P(a) + P(b) - P(a)P(b). Or assuming both drives are identical, 2P-P^2....which, for very small failure rates, is slightly less than double.

I'm not disagreeing that Raid 0 is, for desktop users, usually a bad idea. Far from it. I'm merely correcting the math.
 
Okay, I agree. Let's work it out, though - just two drives, ignore the controller - 2P-P^2 for drives with a MTBF of 500k hours.
P(2P-P^2) = ((2/500k) - (1/500k)^2) = 0.000003999996
So instead of reducing the system MTBF to 250,000 as expected, we get... 250000.25. See why I'm not concerned with this? Your figures are different from mine by a ten thousandth of a percent. I could work out the whole 5-component system, given enough time (it's quite an expansion - there are what, 37 terms?), but I don't think it's worth quibbling about. It's probably within, say, one percent of what I said, right? Within the ballpark, anyways ;)

 
unhappy_mage said:
Okay, I agree...Your figures are different from mine by a ten thousandth of a percent.
Click to expand...
In the mathematical world, thats the difference between the Nobel Prize and teaching the 8am freshman calculus class :p

A small difference, yes...but for large failure rates or for a very large number of items, its not insignificant. Just something to remember if you ever want to calculate for anything other than small raid arrays.
 
Just run JBOD, you would still see that one nice big logical drive but you wont have the failure associated with raid0, so if a drive fails, you only lose what's on that drive.
 
masher said:
In the mathematical world, thats the difference between the Nobel Prize and teaching the 8am freshman calculus class :p
Click to expand...
Well in that case, it's actually a difference of .[overstrike]000001[/overstrike] :p

 
whoa. too many numbers. to put it bluntly, is it a difference in being screwed, or REALLY being screwed screwed if i used RAID 0? ha ha

and how does JBOD work? like, if RAID 0 strips, how does JBOD span it's files over multiple drives?
 
You'd be slightly screwed if you lost a disk in a JBOD configuration - all the data on that disk would be gone, and the file system is going to complain a lot if you try to copy things off that part of the volume. However, with raid 0 you're completely screwed - all the data on all the disks is then unreadable.

JBOD works by concatenating the disks. If you read through the logical volume in order, you'd get all the bytes from disk 1, then all from 2, then 3, etc. Compare this with raid 0, where you get a small chunk of 1, then a small chunk of 2, then 3, and once you've got to the last disk you start back at 1 again. JBOD doesn't help sequential transfer rates as compared to a single disk, but it doesn't hurt them, either, and it maintains the drive's assumptions about how it should buffer data, leading to possibly better performance in those areas.

 
Everfalling said:
whoa. too many numbers. to put it bluntly, is it a difference in being screwed, or REALLY being screwed screwed if i used RAID 0?

and how does JBOD work? like, if RAID 0 strips, how does JBOD span it's files over multiple drives?
Click to expand...


JBOD:
Say each drive is like a bucket and the data is water, it would simply flow over to the next drive. There is no redundancy, but at least if you lose only one drive, it's only one drives worth of data... not the entire arrays data.
 
Gee, sounds like my Discrete Math class with all the talk of the probabilities :( JBOD is nice, because it basically takes all your drives into what the OS sees as a giant drive (c'mon, is it not cool to see My computer report "1.2TB" :p), whilst basically keeping the same performance/data loss effects as 4 separate drives.
 
ok... so... RAID 0 is bad for small files, doesn't increase proformance unless for larger files, has a chance of failing like, to the power of 4 more than usual (or some other big number) and if it does fail, trags down everything with it? hmm... and JBOD doesn't do much better, though if one drive fails, it might just take down just its own data, but it'll still screw things up, and evidently copying stuff back and forth is a hassle... well shoot.. so much for my day dreams of one large terabyte drive >>
 
Everfalling said:
well shoot.. so much for my day dreams of one large terabyte drive >>
Click to expand...
If you want a very large, secure drive array, you can always go with Raid 5. Of course, its write performance is pretty abysmal, but your data is safer there than it is on even a single drive.
 
RAID 5 is da bomb, yo. Yeah, writes are a little slower, pah, whatever, you write once, you read over and over again. Reads can be slightly sped up similar to RAID 0, *and* you have recoverability if one of your drives fails. How sweet is that?!

For the OP - RAID 5 takes at least 3 disks, but can use as many disks as you like. It's similar to RAID 1 in that you lose a disk of space for recovery information, but unlike RAID 1, it's always just one disk, as opposed to half the space (so if you have 5 disks, you still only have one "used up").

Basically, the way it works is that whenever you write a file to the disk, the RAID controller also writes verification info to another disk. Using that info, it can rebuild any one disk if that disk should fail. It's not quite as good as RAID 1 where you just have all that data still, but you 100% can get it back after drive failure by rebuilding the lost disk.

The only caveat about RAID 5 is that all the disks have to be the same size (and ideally should be exactly the same model). If you have differing size disks, it just uses an amount from each equal to the smallest disk in the array. Luckily, this won't be a problem for you since you bought 4 identical disks.

So for your 4 300 gig disks you'd only get 900 gigs of space, however, you'd be secure against disk failure, which is generally worth losing a disk in my opinion.

-Nate
 
RAID5 isn't bullet proof as some are lead to believe. It's a safer alternative, but they are known to fail and also the task of finding the bad drive might be a lot of work too unless if you had a hotswap cage that would indicated a defective drive.
 
Ockie said:
RAID5 isn't bullet proof as some are lead to believe. It's a safer alternative, but they are known to fail and also the task of finding the bad drive might be a lot of work too unless if you had a hotswap cage that would indicated a defective drive.
Click to expand...

It's true you can have 2 drives fail at the same time and lose your Raid5, and its even true that a Raid5 array may get corrupted. But the chances of both are EXTREMELY low. I've been in IT for 19 years and Hobbyst for 22. In all that time, I've seen one Raid5 actually fail. So in realm of possibilities, anything is possible, but not very likely.

As for identifying a bad drive, honestly....come-on! Its not that hard, it may take 10-15 mins depending how many drives you have, but its not that big of a deal. Many newer raid cards even have lights that will indicate which drive is bad...
 
Ockie said:
RAID5 isn't bullet proof as some are lead to believe. It's a safer alternative, but they are known to fail and also the task of finding the bad drive might be a lot of work too unless if you had a hotswap cage that would indicated a defective drive.
Click to expand...


Well, there isn't anything that is bullet proof. Let me restate that: there is NOTHING that is bullet proof.

Yes, two drives can fail within a short period of time, but as with all data protection schemes we are playing a probabilities game. In the end, people try to reduce the data loss probability. I understand your point about Raid-5 being more complex than JBOD, thereby being more error prone, however R-5 has other strenghts.
 
Ockie said:
RAID5 isn't bullet proof as some are lead to believe.
Click to expand...
At my lab, we have just over 150 Raid-5 arrays, each containing 14 disks, and each running 24 hours/day for the past 7 years. Plenty of disks have failed...but we haven't lost a single byte yet.

The real dangers to Raid 5 arrays are _external_. If you have a software flaw that writes bad data into the array-- you lose the data. If someone nukes your site-- you lose the data. Otherwise, as long as you rebuild as soon as any disk fails, your risks are nearly negligible.
 
so raid 5 requires i have at least 3 drives, but one of them will always be full to be used as backup for the others? hmm.. so my 4 drives will be turned into 3? hmm.. well i suppose that's not too bad...
 
Everfalling said:
so raid 5 requires i have at least 3 drives, but one of them will always be full to be used as backup for the others? hmm.. so my 4 drives will be turned into 3? hmm.. well i suppose that's not too bad...
Click to expand...
You are correct on the bottom line, you will lose 1 drive's worth of stroage, however there is no dedicated 'backup' drive, but rather a partity scheme that is used so that you can rebuild the missing data later.

May I point you to this SR Article covering a lot of basic RAID info. Maybe you should read that first, then come back with specific questions that the article does not cover.
 
Everfalling said:
so raid 5 requires i have at least 3 drives, but one of them will always be full to be used as backup for the others? hmm.. so my 4 drives will be turned into 3? hmm.. well i suppose that's not too bad...
Click to expand...

In raid5 you lose 1 drive to the parity, no matter how many drives you have. I have 8, and still only lose one. Raid6 allows you to use 2 drives for parity, those improving reliability over raid5.
 
ok, assuming i buy 5 drives, i now need a controler card, because my a8n-sli delux only has enough ports for 4 drives at a time. so what would be a good controller card with 5-6 sata150 ports? preferably something in pci. also, i obviously dont wanna have to spend a whole lotta money.
 
The highpoint 2220a is always my suggestion for small raid arrays where speed isn't a huge priority. It'll work on regular pci slots despite the fact that it's pci-x, it supports 8 drives, it's generally a Good Thing. Around $250 but worth the money.

 
Everfalling said:
ok, assuming i buy 5 drives, i now need a controler card, because my a8n-sli delux only has enough ports for 4 drives at a time. so what would be a good controller card with 5-6 sata150 ports? preferably something in pci. also, i obviously dont wanna have to spend a whole lotta money.
Click to expand...
I believe that your board has a total of 8 SATA ports. If you do software raid, it is possible to have the drives attached to different controllers.

If you want hardware RAID-5, you will be spending some money.
 
the 8 sata ports are controlled by two pices of software. 4 of them are using some nvidia controller, the other 4 are using something else i can't recall at the moment. is it possible to run them both for the same raid? also, if i got 5 drives for storage, there's the matter of having a 6th already being used as my C drive. So that means, let's say 4 of the 300 gig maxtors are one the nvidia controller, one more maxtor is spilled over to the other set of sata's, and then one WD is on the second set of 4 sata's that wont be included in the raid. is this possible?

kinda like this:

1st set of sata ports:
[max] [max]
[max] [max]
2nd set of sata ports:
[max] [WD]
[xxx] [xxx]

where as all the 'max' drives are on Raid5, while the WD is left alone to be it's own seperate C drive.

does software raid setups work the same or worse than hardware ones? and is this setup even possible? AND, if possible, which set of 4 sata ports should i have the majority of my drives on (i'm thinking of putting the 'spill over' drive as my parity one, so which software controller would be best for that?).
 
You must log in or register to reply here.
Back
Top