Slashdot Mirror

I have "lost" a disk in my RAID5, and the NAS rebuilt easily once I replaced it - but if you lose a disk in RAID0, then wave goodbye to your data.

With modern drives sized in terabytes, the same thing can happen to RAID5 if you lose a disk while rebuilding.

Hence the other part of my comment which you didn't quote: "and if I was running in an enterprise with a DS1512+ then I would certainly consider running in RAID6" (although at the time of writing I was actually thinking of the DS1812+ which is more suited to RAID6 as it holds more disks)

I have seen a RAID5 fail whilst waiting for a vendor to replace a disk in a corporate environment - let's just say the vendor had some explaining to do...

-- Pete.

I have "lost" a disk in my RAID5, and the NAS rebuilt easily once I replaced it - but if you lose a disk in RAID0, then wave goodbye to your data.

With modern drives sized in terabytes, the same thing can happen to RAID5 if you lose a disk while rebuilding.

That reminds me of a double WD hard drives failure within a week, the main HD and its backup (this happened in a cloudless life). Amazing, they're even synchronized...

This is totally f@#$ standard. There are two drives bought at the same shop at the same time. Do you think the manufacturers specially made sure to mix them all in with drives from different places? They aren't even just the same batch. They have probably been produced within seconds of each other.

What do you think happens when a drive fails? Some capacitor has been made with the wrong chemicals; some piece of metal has impurities. Some bit was screwed too tight and is weakening the rest of the structure. It's not deliberate; it's may not even be outside normal manufacturer's specifications, but it's the thing which ends up as the weakest link in your drive. The drives made just before have the same metal; the same capacitors and have been put together by the same machine which is reaching the point where it needs calibration. They are all going to fail at the same time.

At the point where the first drive fails, the second one suddenly gets more load (e.g. double the number of reads; up to N-1 times the number of reads if it's part of a RAID array before you even take into account that some idiot normally rebuilds the array before backing it up). It's extremely close to failing as it is. Rebuilding is likely to drive it over the edge and kill the array.

Don't talk to me about the idiots at HP, IBM and every other bloody server manufacturer who put a series of ten identical drives with consecutive serial numbers in their RAID arrays and then sell it as ...

For more start here; and this also seems worth reading

http://www.miracleas.com/BAARF/BAARF2.html

Never use RAID5! Here's the conclusion from the following article:

"For safety and performance favor RAID10 first, RAID3 second,
RAID4 third, and RAID5 last! The original reason for the RAID2-5 specs
was that the high cost of disks was making RAID1, mirroring, impractical.
That is no longer the case! Drives are commodity priced, even the biggest
fastest drives are cheaper in absolute dollars than drives were then and
cost per MB is a tiny fraction of what it was."

With 15 1.5TB drives in a RAID6, I wouldn't bother. It's just too likely that one or more drives will not merely fail but return garbage data, which RAID3/4/5/6 will propagate across the array on rebuild.

If you do this, don't focus on building systems redundant. Focus on building redundant systems. Make your software tolerate losing any one array or server. If you lose an array, don't rebuild; start again from scratch.

Also, it's quite clear that these servers are not particularly speed-oriented. They have 15 drives serviced by a single PCI SATA controller (through SATA port multipliers). PCI's maximum bandwidth is 133MB/sec (33MHz, 4 bytes per cycle). Today, two drives could soak that with sequential reads and writes. That's fine for this company's users (who are backing up to these servers over their slow internet link) but if you're thinking of putting one in at your office to back up your 20TB fileserver, you better think long and hard about how much time you'll have to do a complete restore from it before you're filing unemployment claims. You could speed things up quite a bit by putting in two PCIe 24-port RAID controllers, dispensing with the port multipliers, and using RAID10, but that does add a few thousand to the price and cut the capacity quite a bit.

The RAID5 is probably planned for longer term bulk storage since it can be a tad slower than Raid1.

RAID 5 isn't worth it. If you want to put four drives in a RAID, use RAID 10. Writes are faster on RAID 10 than on RAID 5, and if two drives fail, there's only a 33 percent chance of needing to restore everything from backup, compared to 100 percent for RAID 5.

Why not run RAID-5 (or 50 or 15) if it is seek-heavy?

Because four drives in a RAID-10 are three times as reliable as the same four drives in a RAID-5. Arrays of large drives are more vulnerable to drive failures during reconstruction than arrays of small drives, and RAID-5 is much more vulnerable to a double drive failure than RAID-10. In RAID-5, you lose data if any two drives fail. In RAID-10, you lose data only if the drives that fail are from the same mirrored pair, and there's only a 1 out of 3 chance that two randomly selected drives will be from the same pair. This makes it that much less likely that you'll have to roll back to the tapes.

That and you don't have to read-modify-write to calculate parity on RAID-10, as XanC pointed out.

That blog post forgets one thing: sector remapping.

You can't remap a sector if an entire drive fails. By the time you have put in your spare and your RAID controller has finished rebuilding the arrays, there's a not insignificant chance that the stress of constant reading has killed one of the remaining good drives. This chance goes up as the time to write an entire drive (capacity divided by transfer rate) goes up. See Art S. Kagel's article that strongly recommends RAID 10 over RAID 5.

Adding to AC.

See this. 'Nuff said.

http://www.miracleas.com/BAARF/

Add an 8 port 3ware SATA Raid controller (configured to RAID5)

I would highly recommend against a RAID-5 configuration. You're not going to save much money over RAID 1+0 and it's a ticking time bomb. See here for a cost analysis, basically you save about 33.3% by going RAID-5, but the write performance is worse than a RAID 1+0 array.

Do the math for a multi-disk failure in a RAID-5 versus a RAID 1+0 system (let's say four disks). Both the RAID-5 and RAID 1+0 system will handle a single drive failure without any problems. Now, see what the odds are of the RAID system going down when a second drive fails; in the RAID-5 case it's 100%, in the RAID 0+1 case, it's only 33.3%. More information about this is available here. I don't know about you, but having to possibly restore a terabyte array from either tape or optical media is not my idea of a good time; spend the extra cash for RAID 1+0.