RAID5 is not easily restorabale?

V

Val

Limp Gawd
Joined
Nov 1, 2005
Messages
299
RAID5 seems to be the best and fast option while still maintaining reliability. However i searched this forum a bit and some people say its not easy to restore RAID5 array.

As far as i understand, when RAID10 fails you just get a message that a hard drive has failed, you buy a new one and put it in place of the broken one. No reinstalls, no downtime, nothing.

Is it same with RAID5 (except that it drastically slows down)? Or will i have to reinstall windows and run a special program?
 
generally raid5 is easy to restore you just add the new hdd and go. and it doesn't have any down time. but it is slower whe it is rebuilding.
 
Forgot to add, the plan is to get P35 motherboard with ICH9R chipset...
 
Adidas4275 said:
generally raid5 is easy to restore you just add the new hdd and go. and it doesn't have any down time. but it is slower whe it is rebuilding.
Click to expand...

Does it apply to all controllers? See above post :)
 
Your controller should support that feature (the ability to hotswap and rebuild) -- otherwise, it's not a good controller. The same thing happens with RAID 10, you just replace it and continue on your business.
 
It works the same in RAID5, only I doubt hot swap is possible on the ICH9.
You'd be notified either by the Intel Matrix RAID Manager software or an event log that a drive is dead, then you'd shut down and replace the drive.

While the array is degraded performance would suffer and the rebuild could take from 1 to several hours depending on the size of the array.
 
No need for hot-swap, just want real-time backup.

Great, thanks! Guess thats the way to go then, since RAID0 is unreliable and RAID10 is slower :cool:
 
RAID 10 is faster. I've read the literature comparing RAID 5 and 10 as well as the tests -- in read, they're not much different, in write, the RAID 10 always comes out on top. In degraded performance, the 10 comes out even more on top.
 
gigabyte1024 said:
RAID is not a substitute for a backup solution.
Click to expand...

QFT. RAID is downtime prevention. It does give you a bit of drive failure protection, but it makes the system more complex and thus more likely to fail. For more than 3 drives, you WILL have two fail close together at some point (...well, for 3, it's just less likely to happen in the first year).
 
ashmedai said:
QFT. RAID is downtime prevention. It does give you a bit of drive failure protection, but it makes the system more complex and thus more likely to fail. For more than 3 drives, you WILL have two fail close together at some point (...well, for 3, it's just less likely to happen in the first year).
Click to expand...

You're right about that... the system does become more complex, however, if your RAID card fails, I'm guessing you will probably be able to simply replace the card and restore 99% of your data.
 
Val said:
No need for hot-swap, just want real-time backup.

Great, thanks! Guess thats the way to go then, since RAID0 is unreliable and RAID10 is slower :cool:
Click to expand...

Where did you get that idea about RAID 1+0? Slower in what sense? RAID 5 random write performance is not as good as a RAID 1+0 because of overhead from parity. I think RAID 5 degradation can also be pretty significant whereas RAID 10 has less of a problem with it. The trade off is that you give up storage efficiency with RAID 10 (only 50%) so it makes it pretty expensive to implement.

I also echo gigabyte1024's advice. RAID is no substitute for backups of your data. It is for reliability and perhaps performance, but backups should be static and unchanging.
 
Cyrilix said:
RAID 10 is faster. I've read the literature comparing RAID 5 and 10 as well as the tests -- in read, they're not much different, in write, the RAID 10 always comes out on top. In degraded performance, the 10 comes out even more on top.
Click to expand...

This statment doesnt make sence if you dont mention controller.
According to this peformance in write is about equal, but there is big difference in read: (for ICH9R)

http://tomshardware.co.uk/2007/05/2...chipsets_with_fsb1333_and_ddr3_uk/page13.html
 
Cyrilix said:
You're right about that... the system does become more complex, however, if your RAID card fails, I'm guessing you will probably be able to simply replace the card and restore 99% of your data.
Click to expand...

The card is fortunately the least likely thing to fail (and you can just swap it out for a new one without rebuilding the array). An interesting bit is that sometimes drives go into a recovery mode on the order of ~2 minutes...but this will usually get it kicked from an array, so you can get good drives being dropped. There are special versions designed for RAID that avoid this. The sorts of things that make you need to run CHKDSK and maybe lose a few files will now nuke your array.
 
RAID5 is slower because it has to calculate parity for writes.
RAID10 or 0+1 (same thing I think, mirroring + striping) doesn't have parity to calculate.

Re: Backups, what will you do if your (mothboard or PSU) fails and takes all the drives out with it?
 
I think performance should be the same or near same between RAID 01 and 10, but you never know with the RAID controllers. I find it interesting/impressive that RAID 5 in the ICHx series has such good write performance, on par to the RAID 01 in the ICHx series (unless the RAID 5 array is degraded). Look at the 680i and that paints an incredibly different performance picture between RAID 5 and RAID 01.

However, RAID 10 is still preferable if you go that route because it has better fault tolerance than RAID 01 (10 is a striped mirror, 01 is a mirrored stripe).

I guess the ICH8/9s just really excel at RAID 5? I haven't really examined it before. :eek:
 
gigabyte1024 said:
RAID5 is slower because it has to calculate parity for writes.
RAID10 or 0+1 (same thing I think, mirroring + striping) doesn't have parity to calculate.
Click to expand...

But why does RAID5 show clear superiority in the above link?

gigabyte1024 said:
Re: Backups, what will you do if your (mothboard or PSU) fails and takes all the drives out with it?
Click to expand...

Thats a good point. Does it really happen often enough on good hardware to worry about it? I have constant problems with hard drives (lost only one so far though, mostly just overheating), but yet to have anything remotly doggy with mb or psu...
 
Val, from your posts you're really concerned with disk and overall performance, you might consider waiting another quarter for the X38 to ship. (or at least until someone comes out with a comparison of the two chipsets)

I still say 10 and 0+1 are the same thing, it's a Striped set with a duplicate set.
Someone have something that says otherwise with link?
 
AT power supplies seemed to 'splode all the time...ATX not so much if you avoid the bad ones.
 
gigabyte1024 said:
Val, from your posts you're really concerned with disk and overall performance, you might consider waiting another quarter for the X38 to ship. (or at least until someone comes out with a comparison of the two chipsets)

I still say 10 and 0+1 are the same thing, it's a Striped set with a duplicate set.
Someone have something that says otherwise with link?
Click to expand...

Conceptually that's it, but implementation wise they're a little different.
http://www.storagereview.com/guide2000/ref/hdd/perf/raid/levels/multLevel01.html

In RAID 0+1 you stripe before mirroring (so you mirror two stripes). In RAID 1+0 you mirror before striping. One disk from each mirror in RAID 1+0 can fail and there is no data loss. This is not true in RAID 0+1, as multiple drive failure is only tolerated only if the drives in one stripe were to fail.
 
Val said:
As far as i understand, when RAID10 fails you just get a message that a hard drive has failed, you buy a new one and put it in place of the broken one. No reinstalls, no downtime, nothing.
Click to expand...

I think the requirements are dependent upon the controller. Some may require you select a drive that can be used to rebuild. Others -like the one on my Abit i90HD- have no options for rebuilding RAID-1 arrays, nor any method to find the array status in windows.

Intel's software is pretty nice. It has a balloon pop-up that notifies the user if the RAID array is degraded. Upon adding a new drive, you can select that as rebuild target.
 
Yep, Intel Matrix definately allows to easily rebuild RAID5.

So what about 10 vs 5? It looks like ICH9R definately favours RAID5, despite that many people say 10 should be faster?
 
Cyrilix said:
Regardless of your controller, I still don't see why RAID 5 would be faster than RAID 10 since they're both striping, but RAID 5 has to do parity work.
Click to expand...

So you saying Toms Hardware is bullshitting us? :)
 
Val said:
So you saying Toms Hardware is bullshitting us? :)
Click to expand...

I'm saying there is a misunderstanding that someone on my level is unable to solve, however, I've seen stuff that goes against what you've showed me, with good explanations for them.

If I had to answer you, then yes, I think either:

1 - Tom's is bullshitting us. :p

2 - Something is wrong with the controller in either software or hardware.
 
Cyrilix said:
Here's a RAID 10 white paper.
Regardless of your controller, I still don't see why RAID 5 would be faster than RAID 10 since they're both striping, but RAID 5 has to do parity work.
Click to expand...

This is only a guess I have nothing to substantiate it: If we assume the ICH9R is very efficiant at RAID5 parity, the performance difference could be the fact that RAID10 has to effectivly write to two arrarys (stipe it, then mirror it) In the time it takes to push that through the controller, it could be writtien once w/ parity to the single RAID5 array.

I didn't read Tom's article, but the number of disks was not apparent on that chart, more disks = faster reads and writes for RAID5.
 
gigabyte1024 said:
This is only a guess I have nothing to substantiate it: If we assume the ICH9R is very efficiant at RAID5 parity, the performance difference could be the fact that RAID10 has to effectivly write to two arrarys (stipe it, then mirror it) In the time it takes to push that through the controller, it could be writtien once w/ parity to the single RAID5 array.

I didn't read Tom's article, but the number of disks was not apparent on that chart, more disks = faster reads and writes for RAID5.
Click to expand...


I only read that chart as well, but lets not talk about writes. On the graph, their writes are about the same, but lets look at the reads. 3 times faster on reads for RAID 5? Yeah right.
 
Damn, turned out to be an interesting thread :) lots of very interesting info for a n00b like me.
 
Cyrilix said:
I only read that chart as well, but lets not talk about writes. On the graph, their writes are about the same, but lets look at the reads. 3 times faster on reads for RAID 5? Yeah right.
Click to expand...

RAID5 reads from more disks than raid 0 or raid 10, I believe those numbers.
More spindles = more speed.

edit: wanna see some big numbers from RAID5? Look here: http://hardforum.com/showthread.php?t=1179305
Granted I don't have apples to apples R5 to R0, but it shows the read power of raid5.
 
gigabyte1024 said:
RAID5 reads from more disks than raid 0 or raid 10, I believe those numbers.
More spindles = more speed.

edit: wanna see some big numbers from RAID5? Look here: http://hardforum.com/showthread.php?t=1179305
Granted I don't have apples to apples R5 to R0, but it shows the read power of raid5.
Click to expand...

I have no evidence on which to disagree with you, so I will have to agree, however, if so, then that Tom's bench is highly ambiguous, and they're probably using something like 6 drives in RAID 5, at least.
 
bashPenguin said:
I don't see RAID 10 in that. Also, is this sequential write, or random write?
Click to expand...

I am having a hard time finding that information. I shall randomly guess and assume that it's HDTach's R/W version.

The next page has some IOMeter benchmarks, which make RAID-5 look a little less `supreme'. The Web server bench has yet a different set of results. In the database pattern RAID-5 is completely whipped into submission by raid 0+1.

The point is that a single benchmark is going to tell us very little about how the harddrive or array will perform. Depending on the OP's usage, different setups are preferential. The easy to measure and easy to quote STR is a nice metric to have, but does not paint the complete picture. Surely, few members on this forum would make a graphics card purchasing decision solely based on pixel fill rate, or would they?
 
you don't even need a swap. hot spare'll work. you can even have multiple hot spares. only downside is they are vunerable as the other disks are to power death, temperature death. You can do a thing on some controllers where it'll shrink the array and rebuild if there's enough free space and create a smaller raid5 and exclude the bad hdd.
 
So, since im not going to be hosting database/file server on my computer, but mainly playing games and edit/rip/record videos, it looks like RAID5 with around 4 drives is a solid choice?

What about loading times? Searched many reviews but didnt find anything on it for RAID5...
 
It depends on the game. If they do sequential reads, sure, it'll be plenty fast. But if they do small reads and writes all over the place (and some games do) it's gonna be slow. Probably slower than a single disk.

Conventional wisdom says raid 5 isn't for games. But if you want to try it and let us know what happens, we'd be glad to listen :cool:
 
You must log in or register to reply here.
Back
Top