• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Is RAID worth it? Please read Tech Report's Article before responding. 50 year olds and older only!!! Unless you consider yourself SMART

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.


<< Abit KT7-RAID (Highpoint 370 ATA100 RAID controller, version 1.03 drivers) >>




look, it's not even hardware IDE RAID. it's software, and it's only ONE controller...

I have never liked the highpoint controllers, which is my own prejudice =P



<< 2x Maxtor 20GB 7200RPM ATA/100 hard drive >>



and i don't have maxtor drives either.
 
They do Agent004 when people are using &quot;averages&quot; and &quot;probability of failure&quot; incorrectly. In one example previously, if two drives failed at 10,000 hours each (ie they both operated perfectly for 10K hours), the average failure rate was 5,000...which is completely wrong.

Eug: I understand exactly what you are saying, but winning a lottery, in this example, has nothing at all to do with the # of tickets purchased. Yes, your chances of winning are increased with every ticket you buy, but your odds of winning on any one ticket remain the same. Your probability of winning, per ticket, does not change regardless of the # of tickets purchased.
 
RAID-0 alone? Hell no. RAID 0+1? Yeah, that's where it's at 🙂

I'm using 4 60GXPs in a RAID 0+1 and I absolutely love it. And there's no worry about data loss.
 
WOW!


This thread is WHACKED out.


I use IDE Raid 0 and it rules. <---- Notice the period

Yeah, It doesn't give me even 1 more FPS (I run a software array by way of an HPT-370 on my KT7, so it probably actually gives me LESS fps), BUT, my games LOAD way faster. People who say there is no point to IDE RAID really need to apply their crooked way of thinking to SCSI RAID as well. Think about it, how can SCSI raid be the ultimate, and IDE raid be useless??? Some people are freakin wierd. IDE RAID is way better than IDE non-RAID. It may not be as good as SCSI RAID, but useless??? I think anyone that says it's useless has probably not tried it, or owns a SCSI setup and just thinks they are the king (That would be an EGO thing btw). The only other splaination I can come up with is that they were looking for a new toy, heard that IDE RAID was useless (Without actually doing any research) and spent their money elsewhere and are now refusing to kick themselves for the bad choice (This is also an EGO thing). I can tell you from experience, IDE RAID is WAY worth it.

All you have to do is be smart about it. Don't install windows to the array, but DO put the swapfile on the array IF it is on a different controller than your windows drive. DO NOT forget that if you lose 1 drive you lose everything (Not that this will happen MORE often, but it's just there is a lot more data involved), so make sure you go through and set all of your office type apps to save and load files from a non-array drive, and make sure that you go through and copy all savegames and stuff to the same. You could also get some sort of backup scheduling program to backup all the important parts of your array to a CD if you have a burner.

I guess I just think it's funny that people cut off their nose to spite their faces. Get IDE RAID if you can, you will thank me, and yes, I did read that article. I liked it except when he said that it was useless in games (Even though he did add &quot;From an FPS point of view&quot😉.
 
In regard to the lottery part, I'll put this really simple so that its a bit more clear.

If your odds are 1 in 16,000,000 of winning...and you buy 2 tickets, you're odds DO not go to 1:8,000,000 or effectively double...here's an example why.

2 of 16,000,000 means there are still 15,999,998 other possible winning tickets (or &quot;chances&quot; of winning or losing for that matter).

1 of 8,000,000 (buying two lottery tickets give you twice as much chance of winning as one) means there are ONLY 7,999,998 other possible winning tickets. Obviously, this doesn't make sense, does it? No, because the probability is NOT the same. But in the way you're thinking, it would be &quot;correct&quot;... except that it's not.

While you are &quot;doubling&quot; your chance of winning, you're not halving the probability of winning (or losing). HUGE difference.

Because probability doesn't work by simple division. Don't feel bad, Travis...for a lot of people, this concept is difficult to grasp.

Is it more understandable now? Again, this is one of the most common misconceptions regarding &quot;odds/probability&quot; there is.
 
Well, your math is correct but your average isn't. If they both failed @ 10,000 hours...then the avg. failure was 10,000 hours.
Drive 1 @ 10,000 + Drive 2 @ 10,000 = 20,000/2 = 10,000 hours. Think about it.

Let's put it this way... you make a system, and drive a fails at 9,900 hrs. You replace it and restore the data, and then drive b fails at 10,000 hours. So once again you replace it and restore the data.

So your system has been up 10,000 hrs and you've had 2 failures, replaced a drive twice, and had to restore your data 2 times. 10,000 hrs system time / 2 failures = 5,000hrs MTBF

Without actually knowing the deviation, all this is moot

the standard deviation of MTBF is a red herring and irrelevant to this theoretical discussion.

The REAL &quot;deception&quot; in the article is that it doesn't make it clear that the failures will occur in sets, and that the MTBF of the raid system will be the same as of a single drive if you replace ALL the drives when ONE fails, like any smart person would do.

(lottery ticket stuff missing...)
1:8,000,000 IS = to 2:16,000,000
No way! Not at all the same thing. Having two tickets does NOT &quot;double&quot; your chance of winning

What you MEAN to say is that &quot;buying 2 tickets does not HALVE your chances of LOSING&quot;.

It DOES double your chance of winning (which is merely 0.00000625% with 1 ticket), and in fact, you could double the # of tickets you buy 18 times (for a total of about 256,000 tickets) before you even have a 1% chance of winning.


 


<< RAID-0 alone? Hell no. RAID 0+1? Yeah, that's where it's at 🙂

I'm using 4 60GXPs in a RAID 0+1 and I absolutely love it. And there's no worry about data loss. >>



I was going to say that is what i would run myself, but you already got that! I would run either raid 1 or raid 0+1 myself. I rather have the redundancy.
 
I wouldn't be all that worried about about data loss with RAID 0. Think about it this way. 2 identical harddrives on the shelf. One has a fault that will cause it to fail in 2 years and 364 days and the other one has a fault that will cause it to die in 3 years. If you go RAID 0 it will fail in 2 years 364 days but if you run a single drive and happen to pick that weaker one its dead anyways. Everyone says &quot;well you double the chance of data loss&quot; but if you run the HD that fails in a single drive setup your screwed anyways. Speak all the #'s and stats you want but in reality the chance of you losing data just comes down to the odds of you buying a bad drive.
 


<< While you are &quot;doubling&quot; your chance of winning, you're not halving the probability of winning (or losing). HUGE difference. >>

😕 Huh? For Aj_UF, you're way off base on the lottery ticket issue. Very simple stats actually. While buying ticket number 2 doesn't make ticket number 1 more likely to win the lottery (in a finite pool of 16 million different tickets), it does make the ticket owner twice as likely to win the lottery.



<< Probability distributions are not geometric. Take a statistics class and learn to calculate probabilities. This site is a basic instruction on probabilities... >>

Rahvin, it's been a while, but I took a couple stats and probabilities classes back in university. Did quite well actually. 😉 However, I think you've missed my point since I wasn't trying to explain MTBF.

In terms of MTBF, I've already said it's quite a different kettle of fish (see below), and that's why I chose not to talk about it. My example of RAID 0 vs. RAID 1 was not using MTBF, but an arbitrary example of probability of failure within a specific period of time. This is WAAAY oversimplified, but guys like me like thinking this way. 🙂

My arbitrary unit is 1/10000 chance of a crash within a specific period of usage, eg. say, 1 year. This is not reality and not directly related to MTBR, but easier to think about for illustrative purposes, just to say that RAID 0 is much less safe than either no RAID or RAID 1.
 
Wow most if not all here know nothing about probabilities!
oh yeah when i roll six times a 6 sided dice my prox for a 1 is sure 1/6*6 =1
We should not forget raid 0 is senseless in these days!
show me a application which needs 60mb/s
typeing this with a 60mb/s raid system 🙂
(and i know it senseless)
raid 1 helps for drive failure but motst crap is done by software!
and there is nothing which can prevent software crashes
 


<< oh yeah when i roll six times a 6 sided dice my prox for a 1 is sure 1/6*6 =1 >>

I hope you're joking because that's even more oversimplified than what I'm saying (and even more incorrect).
 
If I was to go out and buy a 60 gig hard drive and use it in one system and and another go out and buy 2 30 gig drives the mean time to the ORIGINAL Data loss would be almost the same but the RAID would fail sooner because the chance of getting a drive which fails below the MTBF would be twice as high or 75% chance. I think this would thus the mean time before data loss would be one standard devination below the MTBF of the none raid drive. The difference in risk of data loss comes After the first drive fails and is replaced with out replacing the otherdrive. The remaining old disk and the non raid disk would have the same MTBF rates.
 
MTBF is calculated by dividing the total number of failures by the total number of operating hours observed by a HD manufacturer in prerelease Q/A testing. If the rating is for 300,000 hours, that means that on average every 300,000 hours of operation for the listed drive a failure will occur. If you have 2 drives operating for 150,000 hours a piece, 3 drives at a 100,000 hours a piece and so on, odds are one of those drives will fail. So theoretically if you have one drive with a rating of 300,000 hours it should last for 34 years. Obviously that won't happen, so here's the catch. MTBF only applies within the service life of the product. Seagate states the service life of their Cheetah at 5 years. So if every five years you replace that one drive with another brand new one, it will theoretically be 34 years before you can expect a failure. Each additional drive you add does not double the odds of failure, though statistically the first one does. The actual increase in failure rate for each drive you add is one over the number drives you have. If you have one drive and add a second your MTBF drops in half going from 300 to 150, though the total operational time for the 2 drives is 300. If you add a third drive your MTBF drops a third from having 2 drives going from 150-100. A fourth drops MTBF a quarter from having 3 drives, 100-75 and so forth.
 
You buy 1 drive and run it in your system, only to have it fail after ONE week.

Or

You buy 2 drives and run them in RAID x in your system. One of the two fails after ONE week.

------
Putting drives in RAID does not reduce/increase your chances of losing data (unless you use mirroring). The drive that failed after ONE week crashed the user's system in both cases. The only way you increase your chances of crashing a system with more than one drive is in the fact that with the more drives you buy, you increase your chances of buying a faulty drive (which will crash your system...).
 
Naughty, you are saying sort of the same thing I did.

These are stooopid extreme examples of course, but it does illustrate the point:

Let's imagine that you have a 1/10000 chance of a drive failing in one week. If you bought 10000 drives in RAID 0, while you don't have a 100% chance of one of those 10000 drives failing in one week, you are more than like likely to get a crash. Unfortunately, a crash of one drive in RAID 0 essentially makes the data on all the drives in the array useless. You would have been much better off just buying one drive, instead of 10000.

On the other hand, if you had 10000 drives in RAID 1, you can pretty much rest assured your data is going to be intact, since you have 9999 backups should one go down, or 9998 backups if two go down, etc.
 
Ah, the statistic debate😀 We all loved it in college didnt we? Your ridding high on your theory until someone else proves that their theory is correct. The you try and defend your theory, but since you actually realize that his theory is correct, you sliently give in. Im not sure who is correct, but it doesnt really matter. The main drawback of RAID 0 is data loss. Can you live with that? The main drawback of RAID 1 is write times?<--not sure though. I hope you all read the article though, and by guessing from all the statistics in here, you guys are SMART😀
 
&quot;Let's imagine that you have a 1/10000 chance of a drive failing in one week.&quot;

We don't have to guess. The statistical failure rate for any number of identical drives over any specified length of time is easily calculated using the following formula:

(length of time) / (MTBF / # of drives) = % failure rate

We'll use Seagate's IDE Barracuda III as an example as neither Maxtor nor IBM have MTBF's on their site.

1 year 500,000hour MTBF

(8760) / (500,000 / 1) = 1.752% chance of failure for 1 drive
(8760) / (500,000 / 2) = 3.504% for 2 drives

1 week

(168) / (500,000 / 1) = .0336% for 1 drive
(168) / (500,000 / 10,000) = 336% for 10,000 drives

What does above 100% failure rate mean? For every 100% one drive statistically is expected to fail. The last example calculated another way. If you have 10,000 drives each operating for 168 hours that's a total of 1,680,000. You would expect one drive to fail for every 500,000 hours of operation according to the MTBF which would equal at least 3 drives.

 
sorry, i'm not 50 yet. oh well. the only reasoning i can even see behind raid 0 is the really high str. otherwise the risk of data loss isn't worth it. like eug said, it's a great way to have cheap redundancy if you mirror it. and the only time i can see str coming into play is either capturing video, or loading large sound files into main memory, or basically anything involving reads and writes on sequential sectors on the hard disk. otherwise real world performance doesn't see that much of a benefit since most disk activity involves random accesses accross the platter, and access times are virtually the same, if not slightly slower in the case of raid 0. and then there's more system overhead to deal with.
 


<<

<< don't suddenly have 5,000 MTBF since you have two. >>



Sure you do. if both drives die at 10,000h, then the average (synonym for &quot;mean&quot😉 failure time for your system is 5,000. (total running time / # of failures = 10,000 / 2 = 5,000)

the lower MTBF failure just means you're going to get dead HDs more often... which makes sense since you have more of them in your system.



<< In the simplist terms, if you have 1 in 16,000,000 of winning the lottery with one ticket, do you have 1 in 8,000,000 if you buy two? No way (now you have 2 in 16,000,000, which still sucks). >>



1:8,000,000 IS = to 2:16,000,000 .. Too bad you don't remember ratios as well as stats =) >>



don' think so: forget pure mathematical methods and get to real life: if both are going to fail once in every 10,000 hours, the very first single one to fail would be the very first for the couple, so if the first failure was going to happen say at 8,500 hours, the same would be for the couple of drives. The second failure would be another kettle of fish, as it would (could) happen sooner than expected if the system was a single-drive one.
 
Besides all the statistics &amp; all this calculating going on here. There's one intrinsicly ironic part in RAID which ppl have missed out so far (or are unaware of).

RAID actually makes your drives MORE likely to fail. The reasn? Think about it - RAID-arrays have a heck of a lot more HDD-activity than a single drive. Also, the more drives you have, the more likely is it that a single one WILL fail.

So, the redundancy side of RAID obviously is supposed to counteract the &quot;bad&quot; part of stressing the drives more (and, IMHO it definately does pay up - the redundancy that is).

I just find it quite amusing that in a way, RAID is somehow trying to battle what it causes itself to a degree. Ironic, isn't it? ;D.

Anyway - hope some ppl appreciate the irony of it as much as I do 🙂.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top