Web Hosting Talk : Web Hosting Main Forums : Colocation and Data Centers : buying a server, some raid 5 questions

[disgust]

my understanding of raid 5 with three disks is as follows:

a data block is written to drive 1, a data block is written to drive 2. this accomplishes the speed increase. then, on disk 3, a parity file is placed.

my question is, what happens if the parity drive fails, NOT one of the data block drives?

also, assuming a drive fails, how do you rebuild things with another drive? and raid requires the disks to be identical, correct? if a disk fails, how do I know I'll be able to get another of the exact same model of drive?

[tnndotnet]

Parity information is stored on each drive.

So a small piece of parity information is written on each drive. So if one fails, you dont lose all of the parity information. This is one of the more popular raid methods.

Also here is a pretty nifty site I found about Raid 5:

http://www.pcguide.com/ref/hdd/perf/...eLevel5-c.html

Dan

[disgust]

thanks for the link. what does this mean though? "Degradation and Rebuilding: Due to distributed parity, degradation can be substantial after a failure and during rebuilding."

does that mean not everything would be able to be recovered?

[lpmusic]

everything will, lest two drives crap out on you, then you bust out that backup we all know you have.

When you are running with a failed drive (or a hot-spare is rebuilding to join the array) you experience slower speeds because the array is having to reconstruct half(?) the data using the parity.

[cwl@apaqdigital]

1. you need a method to get notification when array member failed, otherwise you won't even know RAID-5 is in degraded state (single drive failed). if 2nd drive also failed in degraded array or before the rebulding is completed, it's toasted! don't think it couldn't happen, the fact is that 2 drives can fail at the same time range if they are coming from the same batch.
2. unless you have standby hot-spare, the RAID5 won't get automatically rebuilt, you need to "manually" tell the raid card to rebuild
3. redundant array only provides redundancy against hardware failures, it can't replace the equally important OS/DATA "backups"

[lpmusic]

Some raid cards do rebuild, for instance the raid cards in most [at least the compaqs i have] compaqs, will automatically rebuild as soon as you pop a good drive into the slot (these are hot-swap drives as well).

As for telling if a drive failed, in free/net bsd i've notice you'll get a load of "soft errors" in /var/log/messages / dmesg

[MrMan]

Newbie questions:
How can you tell which drive failed?

When doing backups, do you backup each drive, since each drive contains different blocks?

[lpmusic]

depending on the setup, in your logs it should tell you which drive, and you should be able to get into the bios for the raid controller and "select" the drive and it'll flash the light on it for you, or the controller may automatically do this.

As for backing up, backup the whole logical drive, just as your OS sees it.

[MrMan]

Oh okay, so when restoring from a backup (hope one never has to do that though), the OS or RAID controller would handle the distributing of blocks to each drive?

[guyellis1988]

Yes - the OS would if its software raid (bad!) or the raid card woulod if its proper raid (good!) you dont worry about the rebuilding just set it going and wait.

We recently had a drive fail on a Raid-5 array (405 gig, 4 x 146gb drives) a nice setup, we were lucky because another drive failed literally 45 minutes after the array had rebuild - if we didnt have a "hot spare" on board we could have been in trouble!

I highly reccomend keeping a hot spare and keeping your OS on a different array

We tend to build OS onto two mirrored 15k 36 gig drives, 4 x 72gb or 4 x 146gb 10k drives for the raid 5 array and a hot spare on the raid 5.

[easyplank]

I have a number of raid5 arrays setup which do have exactly that, hot swap and with the OS partition seperate from the storage arrays as that makes it easier to manage when things do fail.

Saying that, rebuilds can be lengthy but are effective if a drive fails and the machine can continue to run once a drive fails without to much hassle in correcting it.

What RAID cards are you guys using for your setups?

[Gigalocker]

I'm using raid-5 on FreeBSD via 3Ware 9xxx-series raid controller. I'm curious if a drive fails, and you have your hotswap setup properly so it rebuilds the data from parity correctly. How would you know that you now have a dead drive and your array is without a hotswap?

[cwl@apaqdigital]

assuming you meant 'hot-spare' instead of 'hot-swap'. if you do have hot-spare pre-assigned in RAID BIOS, then the RAID card will automatically "deactivate" the defective member, swap in the hot-spare drive, then it will automatically rebuild the RAID5. all these steps are done without human intervention.

all major RAID cards have bundled array management software which provides "event notification' (such as array member fails) to preset email addresses (SNMP forwarding). however, it requires X-windows to work if you use linux/BSD.