What's the problem with RAID 0?

Silencing hard drives, optical drives and other storage devices

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

wim
Posts: 777
Joined: Wed Apr 28, 2004 5:16 am
Location: canberra, australia

Post by wim » Mon Aug 22, 2005 9:40 pm

mattthemuppet wrote:i think the probality of one person being wrong about probabilities =

(n + q + z)/r,

where n = the IQ of the proposer, q = the age they dropped out of school, z = the no. of humanities courses they studied and r = the no. of people talking about probabilities.

I think that's pretty clear :)
:lol: .. shouldn't it be more like
(r + q + z)/n
or, my postulate:
r/(r+1)

mathias
Posts: 2057
Joined: Sun Jul 11, 2004 3:58 pm
Location: Toronto
Contact:

Post by mathias » Tue Aug 23, 2005 8:24 am

where A is the average grade that someone got in math
and B is the overall IQ of that person, converted to a percentile (ie 95% or .95 if smarter that 95% of people)
and C is the visuo spatial IQ of that person, also converted to a percentile
and D is the amount of stimulants that person has in their body, with 0 being none and 1 being a lethal dose
and E is number of times that person has complained in the past year about there being too much math in school
and F is how loud that person's computer is, expressed as a percentile(.75 being louder than 75% of computers)

( (1-A)^0.5 )( (1-B)^0.2 )( (1-C)^0.7 )( (0.3+0.7D)^0.6 )( 1 - 1/(4+3E) )( F^3 )

is the probability of making that person's head hurt with a big equation and

( (1-A)^0.5 )( (1-B)^2.5 )( (1-C)^0.7 )( (0.3+0.7D)^0.2 )( 1 - 1/(2+8E) )( F^5 )

is the probability of making them say something along the lines of "stop using all dem math-things" or "<slopily made up equation> is the probability that you suck".

And next time I'll use boolean.

mattthemuppet
*Lifetime Patron*
Posts: 618
Joined: Mon May 23, 2005 7:05 am
Location: State College, PA

Post by mattthemuppet » Tue Aug 23, 2005 10:03 am

there's a high probability that you're both probably right!

xarope
Posts: 97
Joined: Sat May 03, 2003 8:16 pm

Post by xarope » Tue Aug 23, 2005 6:08 pm

here's my contribution, I'd like to call this my general theorem of probability:

p=1/n

where:
p=probability of concurrence on calculation of probability
n=number of people involved in the discussion

:D

I'll expound on my special theorem of probability at a later date... I've run out of variables to use...

wim
Posts: 777
Joined: Wed Apr 28, 2004 5:16 am
Location: canberra, australia

Post by wim » Tue Aug 23, 2005 10:56 pm

the special theory comes first, and is a subset of the general :P

Bxwrapper
Posts: 21
Joined: Wed Mar 05, 2003 5:13 am

Post by Bxwrapper » Tue Aug 23, 2005 11:39 pm

Just to point out practical usage of reliability calculation:


RAID0
-----
In this setup, all the drives (components) shall be working to in order to work.
Using reliability block diagram, they all will be in series. To get the total probability of failure for RAID0, basically you add all probability of failure for each drive. Example:

If probability of failure single disk is 0.01, for RAID0 with 4 Hard Disks will be:

0.01 + 0.01 + 0.01 + 0.01 = 0.04

It means your chance to have failure is increased by 400% compared to a single disk.


RAID1
-----
In this setup, any of drives will make the system working. In reliability block diagram they are in paralell. To get the total probability of failure, basically you can multiply the probability of failure each drive. Example:

If probability of failure single disk is 0.01, for RAID1 with 4 Hard Disks will be:

0.01 * 0.01 * 0.01 * 0.01 = 0.00000001

It means your chance to have failure is decreased by 1,000,000 % compared to a single disk.

mathias
Posts: 2057
Joined: Sun Jul 11, 2004 3:58 pm
Location: Toronto
Contact:

Post by mathias » Wed Aug 24, 2005 10:47 am

Bxwrapper wrote:If probability of failure single disk is 0.01, for RAID1 with 4 Hard Disks will be:

0.01 * 0.01 * 0.01 * 0.01 = 0.00000001

It means your chance to have failure is decreased by 1,000,000 % compared to a single disk.
And here factoring in the possibility of elecrocution is even more important. Even if it's only 0.001, here that would make a lot of difference.

lenny
Patron of SPCR
Posts: 1642
Joined: Wed May 28, 2003 10:50 am
Location: Somewhere out there

Post by lenny » Wed Aug 24, 2005 11:19 am

Bxwrapper wrote:Just to point out practical usage of reliability calculation:

RAID0
-----
In this setup, all the drives (components) shall be working to in order to work.
Using reliability block diagram, they all will be in series. To get the total probability of failure for RAID0, basically you add all probability of failure for each drive.

RAID1
-----
In this setup, any of drives will make the system working. In reliability block diagram they are in paralell. To get the total probability of failure, basically you can multiply the probability of failure each drive.
Are we back to this again? RAID 1 calculation is correct, RAID 0 is not. As have been pointed out earlier, failure for RAID 0 is when 1 or more drives fail, which is more easily calculated as 1 - (probability of no drives failing). You don't add probability like that, or you will end up with a probability > 1 which only makes sense to management drones.

Bxwrapper
Posts: 21
Joined: Wed Mar 05, 2003 5:13 am

Post by Bxwrapper » Wed Aug 24, 2005 1:50 pm

lenny wrote: RAID 1 calculation is correct, RAID 0 is not. As have been pointed out earlier, failure for RAID 0 is when 1 or more drives fail, which is more easily calculated as 1 - (probability of no drives failing). You don't add probability like that, or you will end up with a probability > 1 which only makes sense to management drones.

As I said it before, it is a sample of practical usage.

Actually, total probabilities for independent events, is more than just addition.

As an example, for 3 independent events are:

P(A or B or C) = P(A) + P(B) + P(C) - P(A * B) - P(A * C) - P(B * C) + P(A * B * C)

So, the result will never > 1.


However, in case RAID0 where Probability of failure << Probability of success, adding the probability of failure of each drive for RAID0 is close enough and easy for everybody to undestand.


Note:

To calculate total probability of 3 or more independents event, can be easier using its compliment:

P(A OR B OR … OR N) = 1 – (1-PA) * (1-PB) * … *(1-PN)

mattthemuppet
*Lifetime Patron*
Posts: 618
Joined: Mon May 23, 2005 7:05 am
Location: State College, PA

Post by mattthemuppet » Wed Aug 24, 2005 3:23 pm

youch, that's some hardcore stats!

Bxwrapper
Posts: 21
Joined: Wed Mar 05, 2003 5:13 am

Post by Bxwrapper » Wed Aug 24, 2005 7:48 pm

mattthemuppet wrote:youch, that's some hardcore stats!
Yes, unfortunately :wink: but doesn't make life much easier.

The simplified version as I pointed out for those 2 cases, have been widely used in day-to-day work for commercial equipment reliability calculation as long as probability of failure << probability of success. And it is always the case.

MoJo-chan
Posts: 167
Joined: Wed Apr 30, 2003 3:49 pm

Post by MoJo-chan » Fri Aug 26, 2005 12:39 pm

I've been using RAID for years. I'm also a computer repair technican, so see a lot of damaged HDs and systems. Here are a few things you should know.

1. Reliability.

These days, HDs rarely just die. Unless it's a power spike (lightning etc), they usually start to get errors first. SMART monitoring will give you advance warning of this. However, if you look after your drives, they will last. Keep them cool, don't kick your PC around, do occasional full drive scans (most RAID software supports this). Make sure you use good IDE cables too.

2. Speed

RAID0 is useful on the desktop. How useful depends on what you are doing. RAID is designed to improve performance in multi-user systems, where there are many users/programs accessing the disks at the same time, and in streaming applications like video editing. The latter case is more likely to be of use to you. I move a lot of big files around and can easily tell when I'm working with the RAID0 array and when it's just the single drive. I multitask quite a bit too, so get some of those parallel access benefits too. Cache memory helps, as data can move at PCI bus speed with no seek delays. Make sure you configure stripe sizes proplerly for your system.

3. Software

Most RAID arrays have a BIOS where you set the array up, and monitoring software for your OS. It is important to know how to use this software, so read the manual. Most arrays save data about each drive onto the drives themselves, so even if you swap cables arround the array will still work. You can even change the controller. Good RAID software is designed for this kind of robustness.

4. RAID0 vs RAID1 vs RAID5

RAID 0 is for pure speed, but some people want more reliability too. RAID1 can offer that to a degree. RAID1 isn't like having a backup - if lightning strikes and your machine burns to the ground, you still loose your data. RAID1 is more about uptime and failing drives. However, it does offer some of the benefits of RAID0, in terms of speed, since all drives can be access in parallel in just the same way. A compromise is RAID10, a mirrored stripe. RAID5 allows for the failure of one drive in the array, at the expense of one drives worth of space. Writes are much slower unless you get a really good hardware RAID (Areca), however.

5. Software vs on-board vs hardware RAID

Software RAID is cheap, but has a few issues. It's not cross-OS compatible. It's a little easier to corrupt (with hardware RAID the OS can't hose the RAID info sector on the drives). It's also slower, and uses more CPU time. On-board RAID is cheap, but most of it is crap. Most don't support RAID5 or 10. Most are little more than software RAID, and are not that fast. Some of the newer ones are better though. Hardware RAID is by far the best way to go. Get a controller from eBay. Performance is in another league, especially if you get one with on-board cache memory (avoid most 3ware cards). If you want RAID5, get one with hardware acceleration for RAID5.

Conclusion

Do your research. Treat your equipment well. Give it a try.

Bxwrapper
Posts: 21
Joined: Wed Mar 05, 2003 5:13 am

Post by Bxwrapper » Sat Aug 27, 2005 12:45 am

MoJo-chan wrote: ......
RAID 0 is for pure speed, but some people want more reliability too. RAID1 can offer that to a degree. RAID1 isn't like having a backup - if lightning strikes and your machine burns to the ground, you still loose your data....
This is called a common caused in reliability enngineering which is very often to be overlooked. Example of this stress are: as mentioned above (very remote), temperature, humidity or a mechanical stress.

In redundancy reliability calculation, this common cause can play significant roles that explains why a plane or even space shuttle were crashed despite multi level of redundancies they have.

mathias
Posts: 2057
Joined: Sun Jul 11, 2004 3:58 pm
Location: Toronto
Contact:

Post by mathias » Sat Aug 27, 2005 8:02 am

MoJo-chan wrote: These days, HDs rarely just die. Unless it's a power spike (lightning etc), they usually start to get errors first.
Thank you for showing why it's important to factor in the possibility of electrocution.

Besides lightning it could also be caused by problems at the power company, a defective power supply, a foreign substance in the power supply, or the power supply malfunctioning because of some other defective components, like faulty plug or socket, defective voltage regulators on the MB or video card, and so on.

MoJo-chan
Posts: 167
Joined: Wed Apr 30, 2003 3:49 pm

Post by MoJo-chan » Sat Aug 27, 2005 3:11 pm

I agree about the power spike problem being a common cause of failure, not to be sniffed at. Every time there is lightning, we get lots of machines with burnt out PSUs and modems. Often mobos die as well, although HDs usually seem to survive somehow.

Anyway, I think the point is this: avoiding RAID for data security is madness. Only a backup will give you any significant extra security.

DVDs are cheap, burners are fast. Consider if you need all those downloaded Family Guy episodes backed up - after all, you can always re-download them.

StarfishChris
Posts: 968
Joined: Fri Jan 07, 2005 7:13 pm
Location: Bristol, UK
Contact:

Post by StarfishChris » Sat Aug 27, 2005 4:44 pm

You can't redownload your digital camera pictures or your work. And as for lightning, don't you use surge protectors (or are they useless)? Surely that would be a much better alternative to replacing a dozen dead PSUs.

As for RAID, you'd be unlucky to have a drive fail within a few years (by the way, here speaks the guy who has never had a hard drive fail, and plenty of old drives...). Restoring data so infrequently is not such a problem that you avoid RAID entirely.
It's coming back

sthayashi
*Lifetime Patron*
Posts: 3214
Joined: Wed Nov 12, 2003 10:06 am
Location: Pittsburgh, PA

Post by sthayashi » Sat Aug 27, 2005 6:06 pm

StarfishChris wrote:And as for lightning, don't you use surge protectors (or are they useless)?
Though this veers on off-topicness, I have to point out that if you value your equipment or data on it, consider getting a Series-Mode surge protector. If you don't know what one is, read this AVS forum post
[size=75][url=http://www.twolf1300.net/sthayashi/SPCR/systems.html]My Power Rig, Storage Rig, HTPC and Main Rig[/url][/size]

mathias
Posts: 2057
Joined: Sun Jul 11, 2004 3:58 pm
Location: Toronto
Contact:

Post by mathias » Sat Aug 27, 2005 6:44 pm

MoJo-chan wrote:DVDs are cheap
Even Taiyo Yudens.

MoJo-chan
Posts: 167
Joined: Wed Apr 30, 2003 3:49 pm

Post by MoJo-chan » Sun Aug 28, 2005 2:18 am

StarfishChris: You'd be mad to keep you digital photos on just your HD. How much does a blank CDR cost? Make two or three for safety. Even then, you need to re-copy every few years. I use PAR2 for extra security.

I work in a computer repair shop, that's why I see loads of dead machines :)

I too have not had any drives fail, despite having at least 20 odd. Okay, perhaps some of the really old ones from the early 90s have failed by now, but I stopped using them before then anyway.

sthayashi: Thanks for the link. You are right about surge protectors - the cheap ones do nothing. Your best bet is to get a small but good quality UPS. No need for more then a few minutes of power from it, the main thing is that the output is totally isolated from the input and thus immune to spikes.

mathias: Yes, Taiyo Yuden are the only brand I use for anything I want to keep. Everything else is a waste of time, far too hit-and-miss. Plus, they burn perfectly at 12x on my drive, better than at 8x in fact.

For those not in the know, the only optical disc still made in Japan are Taiyo Yuden's. TY invented the original CDR, and their discs are second to none in terms of quality and reliability. Many discs use the TY media code, but have inferior dyes and don't last.

One easy way to tell if your disc is cheap is to burn at a high speed (8x minimum) and then run Nero DriveSpeed or similar. If the speed drops off towards the end of the disc, it's because it's poor quality and the DVD drive has to slow down to accurately read it.

plonkersaurus
Posts: 12
Joined: Tue Sep 06, 2005 8:59 am

Post by plonkersaurus » Fri Sep 09, 2005 7:26 am

Bxwrapper wrote:Just to point out practical usage of reliability calculation:


RAID0
-----
In this setup, all the drives (components) shall be working to in order to work.
Using reliability block diagram, they all will be in series. To get the total probability of failure for RAID0, basically you add all probability of failure for each drive. Example:

If probability of failure single disk is 0.01, for RAID0 with 4 Hard Disks will be:

0.01 + 0.01 + 0.01 + 0.01 = 0.04

It means your chance to have failure is increased by 400% compared to a single disk.


RAID1
-----
In this setup, any of drives will make the system working. In reliability block diagram they are in paralell. To get the total probability of failure, basically you can multiply the probability of failure each drive. Example:

If probability of failure single disk is 0.01, for RAID1 with 4 Hard Disks will be:

0.01 * 0.01 * 0.01 * 0.01 = 0.00000001

It means your chance to have failure is decreased by 1,000,000 % compared to a single disk.

Would you not have to take into account that the hard drives in RAID0 would be used less than if it was 1 big single drive?

That would change the probability somewhat.

It really depends if hard disk usage is related to hard disk failures, or if just having the hard drive on would cause that.

However, I use EMU ProteusX which hard disk streams, and RAID 0 would apparently help alot for that sort of application, So i am getting 2 small 80gb hard drives to have programs and non critical data on running in raid 0, and having everything else on a 200gb drive.

Post Reply