RAID 5 Drive Failures 6

[adamroof]

Hey, just a quicky, in a 4 drive(36gb per) Raid 5 array,
What would happen if one drive failed?

The current layout is two partitions on this array, the C: drive and an H: drive

drive C is an 8gb partition with the 2003 OS
drive H is a 94GB partition with user stuff and shares

how does that affect the failure?
there are no other drives on this system.

 

[58sniper]

If one drive fails, you simply replace the drive with a new one, and life goes on. If they're hot-swappable, there is no downtime (typically). Even if both of your drives are on that array, as long as only one drive has failed, you're fine.

The problem arises if a second drive fails before the first is replaced. Then you've got a problem, and would have to restore from tape. In your scenario, if both C: and H: are on the same array, you'd have to replace the drives, then reload the OS and restore from tape per MS's disaster recovery guidelines.

Pat Richard, MCSE(2) MCSA:Messaging, CNA(2)

 

[adamroof]

Thank you.

They are not hot-swappable, and trying to spend 1600 bucks in justification here to make them hot swappable. (68pin drives have to be replaced with 80s)

So my situation with 2 partitions on 1 array with the OS is not that bad for single drive failures? Is it not "smart" though? Shutting down wont "aggrevate" any failures to other drives?

I guess im looking for logic in ordering 6 drives, 2 mirrored for C in one array, and 4 back in a raid 5 on the second array, would that be "better

 

[58sniper]

Yes, that would be better. If one of the drives fails in either the mirror set or the stripe set, you're still going to be ok.

Pat Richard, MCSE(2) MCSA:Messaging, CNA(2)

 

[ShackDaddy]

Having them not be hot-swappable just means you have to take the system off-line to replace the bad drive. The cost analysis should be about whether $1600 is a good investment not to have to have your system offline for 25 minutes or so on an evening that you do the replacement on, and the hassle of having to wait until after-hours.

Having both partitions on 1 array isn't bad for single-drive failures. It just makes things more complicated when the whole thing goes bad (ie, you lose two drives at once or your disk controller crashes.) Then you have to rebuild both the OS and the data, and that takes up substantially more time than only having to deal with one of them.

Normally I do what you are hoping to do: mirror my OS and do a 3-to-16 disk RAID 5 set for my data, depending on how much data there is and how much speed I need in accessing it. That way I can handle OS disk issues separately and if I have problems with my data RAID, I can at least deal with it while my OS partition is still healthy and usable. This is probably what you are thinking about.

Other reasons to do that: isolating OS and page-file activity from your data partition should speed up data access significantly. Over time you also will find 8gb a very cramped area in which to run an OS and install some of your programs.

ShackDaddy

 

[fs483]

One question, during the time that one of the Raid5 drive is down, is the data accessible ? I understand that once the drive is replaced, the data is rebuilt.

 

[adamroof]

i would assume so according to the answers here, ive had raid 5 failures before, but never on an OS in a partition like i have now. Im Not certain if it depends on which drive failed, how accessible it is, hence my initial concerns!

Yes shack, 8gb is tight, i wasnt the lucky person to set this up 2 years ago, 2g free now, so pushin it. 2 more service packs and im done for!

Great comment on the data access, i guess my options are purchase 2 more drives ($400) and do whats "better", but im still in a non-hot situation. We have 4 hour warranty for 1 more year, but by the time the drive gets here...

25 minutes?? Dang you are fast!

 

[ShackDaddy]

The whole point of RAID 5 is to still have your data online even if one drive fails. Most data centers build their arrays using either RAID 5 with hotswappable drives (slight performance slowdown if a drive fails, but the bad drive can be popped out and replaced with a new one, which regenerates on the fly) or RAID 0+1.

RAID 0+1 is when you have two big stripe sets without parity (RAID 0) mirrored to each other. That way you get all the speed of a big no-parity stripe, but with the full protection of a mirror. It's very expensive in disks, but it is optimized for speed.

More than you'd ever want to know about RAID can be found here:

http://en.wikipedia.org/wiki/Redundant_array_of_independent_disks

Shackdaddy

 

[ScottCr]

or instead of going Hot Swappable you should be able to add a hot-spare.

Then in the event you lose a drive, you wouldn't need to bring the system down.

http://scottcross.blogspot.com

Windows and NT Admin.

 

[adamroof]

ok scott, since you brought it up...

What mode is the hot-spare an active partner, not just a dormant spinning drive. Im afraid my controller wont support it(active mode)

Ive heard not so good idea because it could take years for it to be used. However, with the tightly woven drives, may not be a bad idea, to be able to postpone replacements until weekends. The system will still have to be brought down eventually to replace the bad drive.

 

[ScottCr]

Well in an ideal world you'd never use it. But the point is it's there should you need it - it's an insurance.

And yeah, you will need to bring the system down to replace the bad drive but you'll be able to choose when you do it.


Which controller do you have?

http://scottcross.blogspot.com

Windows and NT Admin.

 

[adamroof]

ok, dont laugh...

PERC 4/SC

 

[technome]

may not be a bad idea, to be able to postpone replacements until weekends"
No you do not want to wait for the weekend to replace drives. When an array is in degraded mode it is vulnerable to block failures on a second drive. The greater the number of blocks or drive capacity size the greater the chance of drives dropping out due to bad blocks. Years ago this was not a major issue as the drives were smaller, with the higher capacity drives this has become a much bigger issue with raid 5. If you have Active directory or extremely critical data, I would go for the hot spare route.

This refers to raid 6, but read on about raid 5, under the subheading "Why raid 6" ...

http://www.intel.com/technology/magazine/computing/RAID-6-0505.htm

........................................
Chernobyl disaster..a must see pictorial

http://www.kiddofspeed.com/default.htm

 

[newBPhoneGuy]

technome i think you misunderstood they were saying that if there is a "hot-spare" configured and one disk went defunct the the hot spare would rebuild and you sould then wait till the weekend to remove the bad disk the only time the sys would be degraded would be once the first disk crashed and while the spare is rebuilding.....A Hot spare is ALWAYS a good idea its always good to be safe. even if its never used at least you know that its there just in case

 

[adamroof]

Most excellent!

Thank you all for your input!

 

[ArmyITGuy]

Keep in mind if this is an OS RAID vs. Hardware RAID, that will impact your recovery. If it's hardware RAID you'll have no problems, however if you use an OS RAID, and the "0" drive fails, you're going to have a hard time rebuilding the RAID.

Practical experience with 2003 standard edition speaking here.

 

[dangermouse68]

To answer whether the data is accessible if 1 drive fails in a RAID 5 config. Yes, it is. The 2 partitions are actually spread across all the drives, not just any 1 or 2 of them.

You'll notice that your 4 - 36 GB drives shows about 108 GB of total usable space, not 144GB like one would think. This is because it uses the space for 1 drive (not a physical drive itself) for parity. This parity is what is used when 1 drive fails.

For more info on parity:

http://www.pcguide.com/ref/hdd/perf/raid/concepts/genParity-c.html

 

[fs483]

I understood that RAID5 allows you to rebuild data to replace the lost drive. But I was wondering if the data was still accessible WHILE THE DEAD DRIVE HADN'T BEEN REPLACED YET. I guess there would be a big performance hit since the data that was on the defect drive would need to be rebuilt dynamically as it's accessed.

 

[adamroof]

and the parity is spread as well in assumption...this is a hardware raid...

a logical partition on a logical array doesnt seem logical!

 

[dangermouse68]

RAID 5 is not the best performance option, so a performance hit should be minimal.

The big question is how long can RAID 5 run while missing one of it's drives? If parity is being written to while all drives are functioning (parity being the drive space of 1 drive) where does it get written to when 1 drive fails? Since the RAID 5 set is now using the parity as the missing drive.

It's best to replace the drive ASAP, hence the hot-spare option. If a drive failed on Monday, I wouldn't want to wait until Saturday to replace it. If the hot-spare is not an option, I get a drive ASAP and replace it at the next opportunity, Monday evening/night.