Raid 5 will have problems after 2009?

P

Pyrolistical

Gawd
Joined
Mar 6, 2003
Messages
737
Ok I posted this link in another thread:

http://blogs.zdnet.com/storage/?p=162

It talks about how using large drives (at least 2 TB) in a raid 5 configuration by the time 2009 comes around the size of the drives and the bit errors means you are very likely to have a read error during a rebuild. (IE. you lose data)

He then also writes how raid 6 will fail in 2019 here:
http://blogs.zdnet.com/storage/?p=805

People are laugh left and right when I posted these links, but why? Do you not understand it does not apply to existing smaller raid 5 setups and only if you use drive larger than 2 TB (hence the 2009)? Or do you just plain disagree with the article? If so where did the author go wrong?
 
Well, that's where the ZFS filesystem is shining due to checksums it offers very good protection against BER when used in RAID0 with copies=2 or any redundant ZFS pool.

Also, using 4K sector drives like the WD EARS series may help with BER, as they have more checksum space available per sector; thus a little more reliable.

Still i agree a plain RAID5 is not as safe as many trust it to be. However, in home situations a single bit flip here and there will probably go unnoticed if these fall in movie or music files for example. If it hits important or binary data or even filesystem metadata that would be very bad.
 
One problem is as stated in article, drives URE rate is <1 in 10^14, and in the future, 2TB drives will hit that easily. Well now its the future, and 2TB drive has a URE of <1 in 10^15.

Also as pointed out by sub.mesa, one big reason for 4K sectors is to deal with errors, read:
http://www.anandtech.com/show/2888
 
Right but both of those articles are completely misleading.

Yes your chances of hitting a URE go up with larger drives. but its still like 1 in 100 trillion bits....which is only 12TB so easily attainable with todays drives. Possible but you have a better chance of winning the PowerBall Grand prize twice, back to back then you do of getting a URE. I like my odds :)

But with a URE you dont lose an entire RAID array like he suggest. Only that one file where URE occured will be damaged. And if you have RAID6 it just reads from the next parity drive.

In no way would a URE cause catastrophic controller failure as he suggest.

BTW you do know that RAID isnt a backup :)
 
This is how I fell about RAID.

RAID is not a backup!

RAID is for uptime or performance and lower cost compared to SSD.

Data will always be backed up on one or more copies.

SSD are out but limited to size and very expensive!

I do run RAID 5 on my NAS and I have had zero problems with rebuilding after a drive failure. I also had zero problems while doing a RAID expansion but it doesn't mean that it can not happen. I am also currently using four 2TB hard drives now.

People will always using RAID 5 or 6 in the future and when SSD drives will be the future as prices come down.
 
Also the RAID 6 being obsolete is even more laughable because for raid 6 to fail you need two URE's on the same 'mapped' sector during a recovery which, basically, wont happen. Two failures on different mapped sectors still have a parity bit available in each case.
 
Pyrolistical said:
Ok I posted this link in another thread:

http://blogs.zdnet.com/storage/?p=162

It talks about how using large drives (at least 2 TB) in a raid 5 configuration by the time 2009 comes around the size of the drives and the bit errors means you are very likely to have a read error during a rebuild. (IE. you lose data)

He then also writes how raid 6 will fail in 2019 here:
http://blogs.zdnet.com/storage/?p=805

People are laugh left and right when I posted these links, but why? Do you not understand it does not apply to existing smaller raid 5 setups and only if you use drive larger than 2 TB (hence the 2009)? Or do you just plain disagree with the article? If so where did the author go wrong?
Click to expand...

I disagree with the article.
After each time this report has come out I went ahead and did a check on all my raid arrays at work and each time I did not see a single URE in any of my dozen arrays. This last time I did this in February. With ~30 TB of raid 5/6 checked I did not have a single URE.
 
On the flip side, Mr. Murphy does have a habit of showing up during degraded states in my experience :) That being said, I still disagree with the doomsday scenario that the articles predict will happen.
 
drescherjm said:
I disagree with the article.
After each time this report has come out I went ahead and did a check on all my raid arrays at work and each time I did not see a single URE in any of my dozen arrays. This last time I did this in February. With ~30 TB of raid 5/6 checked I did not have a single URE.
Click to expand...

Do you constantly read all the contents of the drive? The premise of the article is a URE will occur during a rebuild (hence you would be reading everything)
 
I did a check that verified the raid on each block so yes it did read every single byte on each drive on all 30TB + parity disks. Some of the arrays were 7 years old while others were a few months.
 
Last edited:
I get what he is saying, but I just don't see it happening unless all of the disks are heavily degraded to the point of data loss. By that point any ways, the RAID array would be unsalvagable, regardless of drive size.
 
It's not that. It's what happens when an array is degraded and then you do a rebuild and while you rebuild one of the remaining disks has a single unreadable sector. In this case some raid controllers or software will kick out the second disk making the raid array unusable. I do not have a big problem with that.

I do have a big problem with the premise that 1 out of every 6 2TB drives that every sector has been written on have an unreadable sector and corruption.

Edit: Reworded.
 
Last edited:
drescherjm said:
It's not that. It's what happens when an array is degraded and then you do a rebuild and while you rebuild one of the remaining disks has a single unreadable sector. In this case most hardware raid will probably kick out the second disk making the raid array unusable. I do not have a big problem with that.
Click to expand...

You should cause thats not what happens.
A decent RAID controller just skips that sector and moves on. You are just left with a file that is damaged.
 
Possible but you have a better chance of winning the PowerBall Grand prize twice, back to back then you do of getting a URE.
Click to expand...

Ha Ha, I guess that I why I have never knowingly had a URE.
 
drescherjm said:
I'd be interested in the guy with the 100TB server thread if he sees any UREs.
Click to expand...


Well, I'm not that guy, but I currently have more than 80 TB spread out over some servers at my home. All servers use RAID 5, all data is backed up, all data verified using checksums.

No bit errors here.
 
The problem isn't so much the fact that the URE rates are approaching actual capacity, it's that the speed at which you can fill the drive is growing way slower than the capacity of the drives.

Back in the day, you could fill your 36GB 15k SCSI drive in about an hour, now you'd be lucky to fill your 2 Tb drive inside a day, and that larger window of potential problems is why RAID 5 will end up dying off.

It doesn't help that flaky drives which were serving data ok will end up dying once you try to hammer them non stop for 18 hours trying to re-sliver the data. It's how I lost my RAID5 array. Turns out 2 year old drives will start to die horribly once you break them out of the rut they've been in.
 
nitrobass24 said:
A decent RAID controller just skips that sector and moves on. You are just left with a file that is damaged.
Click to expand...

That shouldn't be the default. It never was with 3ware cards.

Using Spin-rite, I was often able to read the data in those sector(s). While it took a bit more time, and I have to take down the array and remove that drive to test and fix the damage, it's arguably better than just living with the damage, considering what might be damaged.

Then again, knowing what was damaged and restoring it is an option (but the card's not going to help you there then, either...), but sometimes things are unreadable at just the wrong spot anyway.

So, again, a decent controller should have the option to skip an unreadable sector on a rebuild, but it should be an option, and not even the default if it is an option.
 
nitrobass24 said:
You should cause thats not what happens.
A decent RAID controller just skips that sector and moves on. You are just left with a file that is damaged.
Click to expand...

not to mention a decent raid controller can be configured for array scrubbing at a scheduled interval so that bad blocks are remapped before a rebuild scenario strikes. it also lets you keep an eye on the health of your drives - and if you see a drive that always has bad blocks during each scrub, you know its time to pull the drive, and again, before a rebuild scenario strikes.

raid5 + scheduled scrubs + backups is perfectly adequate for many scenarios. i've actually gone back to R5 for some of my arrays that backup/duplicate to secondary storage. other arrays are still too big to duplicate files for and hold less critical data anyway (tv shows, bluray/dvd rips) - and for them I go raid6.

on a separate note the author of the article linked in the OP is a moron.
 
Last edited:
odditory: do you happen to know which controllers work the way you stated to correct BER? I'm not even sure ZFS the "write-to-bad-sector" we discussed earlier; though that's a good occasion to ask on the mailinglist. Might just do that since i'm pretty interested myself. :)

I would think all the onboard and software RAID5 on Windows does not correct BER errors as they usually skip the parity when reading. Only if it gets a timeout or I/O error (TLER) when encountering a bit error will it react in some way; until then the damage is undetected. Or when you do a (scheduled) rebuild of course.

The best solution to BER would be to use checksums on the files, so you can tell whether data is corrupt or not. But, for this to work sexy you need a filesystem+RAID engine in one, like ZFS. This allows for access to the "hidden" redundant data directly. Its just a shame those filesystems are not available on Windows.
 
I hope the kFreeBSD project (which makes Ubuntu use the FreeBSD kernel instead of Linux kernel) will bring "real" ZFS to Ubuntu; that would be kick ass for alot of mainstream users if it had some GUI to manage ZFS as well. Time will tell. :)
 
You must log in or register to reply here.
Back
Top