RAID 0, and other stuff

[Mr_Flibbles]

Howdy!

I've been reading some stuff on this site about RAID, and it doesn't seem to condone the use of RAID mode 0. Well, I'd really love the performance boost for my video editing that RAID 0 provides. Is it really that unsafe to use striping?

Also, I've got two drives right now, both with information on 'em: a 30GB 5400RPM, and a 20GB 7200RPM. I'd like to buy an additional 80GB 7200 and use it with my 20GB for video editing, while keeping my 30GB reserved for apps.

Which brings up my final two questions (sorry!): Is it necessary to wipe a harddrive before loading it into a RAID? And what's a good, value-priced 80GB 7200RPM hard drive?

Thanks for listening.

 

[unoc]

The probability to have an unrecoverable error in a RAID 0 array is twice that of a single drive, just as for the performances boost.
When you build a RAID 0 array with two drives, the total capacity is equal to 2x the capacity of the smaller drive. The performances too, will be determined by the slowest drive. For this reason it is really a rule to use identical drives when building a RAID 0 volume.
Yes, It is necessary to have a blank unformatted drive.
80 GB 7200 rpm are today standard values. It is also important to know that not all 7200 rpm drives perform the same when connected in a RAID 0 array. Moreover you must pay attention to Mobo setting, which may influence the PCI bandwidth and consequently the performances of your RAID array.

the last is in the past

 

[flamethrower205]

True what u have said. The maxtor D740X's are cheap yet really great performers in raid.

dusijtpmo- don't use stupid internet jibberish to piss me off

 

[Napoleon]

The probability to have an unrecoverable error in a RAID 0 array is twice that of a single drive, just as for...

Umm, I'd like to challenge that reliability statement. Assuming that the failure models aren't co-dependent, IMO chances of <i>no</i> failure in RAID0 would follow an exponential model, that is to say (1-p)^n, where p is the failure probability of a single drive and n is the number of drives in RAID0.


___
<font color=green><i>Feet are frozen, hair's on fire, on the average, everything's fine.</i></font color=green><P ID="edit"><FONT SIZE=-1><EM>Edited by Napoleon on 11/01/02 03:24 AM.</EM></FONT></P>

 

[HammerBot]

According to <A HREF="http://www.overclockers.com.au/techstuff/a_hdd_shootout/" target="_new">this</A>, the Maxtor D740X perform very poorly in RAID0, compared to IBM and WD

<i><b>Engineering is the fine art of making what you want from things you can get</b></i>
<A HREF="http://www.btvillarin.com/phpBB/viewtopic.php?t=655" target="_new">My systems</A>

 

[HammerBot]

Correct. Check also this <A HREF="http://forumz.tomshardware.com/hardware/modules.php?name=Forums&file=viewtopic&p=55643#55643" target="_new">thread</A>, that has a bit more about RAID0 vs. RAID1 failure probabilities.

<i><b>Engineering is the fine art of making what you want from things you can get</b></i>
<A HREF="http://www.btvillarin.com/phpBB/viewtopic.php?t=655" target="_new">My systems</A>

 

[unoc]

I tested only two Maxtor 740D 60GB in Raid 0 (120 GB total) and two IBM 120GXP 120 GB (240 GB total).
The configuration test is the following
Raid controller Promise Fasttrack 100 TX2
CPU Athlon XP 1800+ @1677 MHz (equivalent to a 2000+)
Mobo Asus A7V333 BIOS ver 1014
FSB 145 MHz
PCI clock 36 MHz
DDR Ram 1024 MB PC2700
FSB / Ram frequency ratio 1/1
OS Win2K SP3
USB 2.0 disabled by jumper on Mobo
Benchmark suite Sisoft Sandra 2002
VIA 4 in 1 drivers 4.43
Raid performance patch (rpp) ver 1.02
Nothing but the OS and SiSoft Sandra was installed on the system.

Two Maxtor on the RAID controller scored 36000
Two IBM on the RAID controller scored 53000

one Maxtor on IDE channel scored 25000
one IBM on IDE channel scored 32000



the last is in the past

 

[unoc]

Please take a look here: http://www.inf.ethz.ch/department/IS/iks/education/PDDBS/Summer02/foils/L4n.pdf

the last is in the past

 

[Napoleon]

Thanks for the link, here's said link in clickable form:
<A HREF="http://www.inf.ethz.ch/department/IS/iks/education/PDDBS/Summer02/foils/L4n.pdf" target="_new">http://www.inf.ethz.ch/department/IS/iks/education/PDDBS/Summer02/foils/L4n.pdf</A>

IMO, this statement there about RAID0 (un)reliability is still false:

if probability of a disk failing is P, the probability of a failure in a disk array with N disks is NxP

That can't be right. If P=0.11 and N=10, the failure probability would be 10*0.11 = 1.1 = 110%...

The probability of failure of a RAID1 array is:

PR1=P1*P2...PN

HammerBot, I'd say that probability of data loss in RAID1 configuration isn't quite as simple as that; there are dependencies between the drives. Let's say you have two drives of same make and model with failure probability P during MTBF period (assume 40000 hours) and you start using them at the same time. If drive1 happens to break after 20000 hours, you still have a very good chance of replacing drive1 and rebuilding the RAID1 array before drive2 breaks, particularly if you refrain from using drive2 until the replacement drive is installed. However, if drive1 breaks after, say, 39999 hours, chances are that drive2 is also on its last legs and chances of rebuilding the RAID1 array are much slimmer. You'd probably need some sort of, whatdoyoucallit, failure probability distribution over time to better evaluate your chances of succesfully rebuilding the RAID1 array.

____
<font color=green><i>Work is the curse of the drinking classes. --- Oscar Wilde</i></font color=green>

 

[HammerBot]

As I wrote in the thread I referred to, N*P is an approximation that is only valid for small values of P.
You may be right that there are dependencies. But that would affect all types of RAID, not only RAID1.
My calculations assume that there are no dependencies.

<i><b>Engineering is the fine art of making what you want from things you can get</b></i>
<A HREF="http://www.btvillarin.com/phpBB/viewtopic.php?t=655" target="_new">My systems</A>

 

[flamethrower205]

Quite odd considering w/ a 2 hdd array of D740x, I get a constant read speed of 80MB/s. That's pretty damn good imo. It's using a Fastrak TX2000.

dusijtpmo- don't use stupid internet jibberish to piss me off

 

[flamethrower205]

Also note they used a highpoint. According to a few benches I read some time ago, the D740X weren't very good w/ it, but they shine w/ a Promise controllers.

dusijtpmo- don't use stupid internet jibberish to piss me off

 

[SolidReactor]

About this mathematicle solution for "failure calculation" in RAID mode 0. (1/x) = that's One in X drivers whitch fails, under a time of useage.

Lets say that there is 4 driver's in RAID mode 0, driver D,R,V and E. and lets say that we use those drivers for a year.
Driver "D" has a 1/1000 (one in a thousen) "fail rate", that means one driver in a thousen fails under a period of time, in this case a year. Driver "R" has 1/1250, "V" at 1/1500 and "E" at 1/2000

this is the calculation:
D + R + V + E = (1/1000 + 1/1250 + 1/1500 + 1/2000) ~ 0.003 ei. 0.3% failure rate.

keep in mind that the numbers are made up but the calculation is rightful.

BUT... there is always a but... there are other variables that must not be forgotten, such as the PSU, Motherboard, Heat/airventilation and more. The real complex calculation COULD be something like this:

1/(X + (t^(1.5+Kt) + MB*n + W + OS)

"X" stands for the manufactory failure konstant, of the drive itself

"t" stands for temperature failure rate, t^(1.5+Kt) is a more complex calculation of the temperature effect on the drive, so keep it low

"MB" stands for the MotherBoard issue, such as the PCI buss, or things connected to it (RAID Controller) ect.. "n" is numbers of drivers.

"W" stands for the amount of watts (power) the system is consuming, don't forget to be careful with the PSU =|

"OS" stands for the "bugg free" Windows =D

Other factors that can not be forgoten is lightning, viruses and USER'S USAGE.

STILL... my opinion is that it's not so dangerus to use RAID 0, just don't forget to backup the importent files

in reply to:
----------------------------------------
That can't be right. If P=0.11 and N=10, the failure probability would be 10*0.11 = 1.1 = 110%...
----------------------------------------

IF the fail rate is "P=0.11", that makes it 11%!! that is 11 failed drivers in 100! even "Noname" Drivers woudn't have that fail rate! (if there where/are any)
yeah.... thanx for reading this I know it's boring =D

b]<i> Solid Reactor© </i></b>