RAID 5 with big SATA disks. Do or don't?

raid hard-drive raid5

9,239

Solution 1

You had me a R5 - don't.

The reason is that in the event of a disk failure you have zero protection until you've replaced the disk and the array has rebuilt.

For large cheapo SATA disks this rebuilt process can take DAYS - meanwhile you are at the mercy of a second disk failing - at which point it's game over.

Also this type of disk is rarely happy to work solidly 24 hours a day and I've seen rebuilds kill disks - again making the whole thing rather dubious.

If you can use RAID 10 over 5 or 6, if you insist on 5/6 then use 'enterprise' disks capable of a 24/365 duty cycle.

Solution 2

Using RAID5 with this configuration, you introduce additional risk to your RAID. With the storage density introduced with modern disks, the likelihood of encountering a bad block on multiple disks is higher.

If using 1TB+ disks, it's recommended to use a RAID6 as opposed to RAID5, as it has an additional parity disk.

If you want greater speed as well as better guaranteed availability, you might consider RAID 1+0 as well.

As far as speed, compare the seek time and various other specifications. Speed does not vary much with commodity disks that are higher density. When I bought disks a few months ago, the largest tier1 SAS disk I could buy was 450GB.

9,239

Author by

Vinko Vrsalovic

A generalist. Or, better put, jack of all trades, master of none. Currently mastering nothing at stackoverflow.

Updated on September 17, 2022

Comments

Vinko Vrsalovic almost 2 years
Is it sane to build a machine with RAID 5 expecting to hold around 5 Tb (and able to grow) with 1.5 or 2 Tb SATA II disks?

It will be used as audio and documents storage. I already have chosen a chassis with 16 3.5" hard drive bays. Now I'm thinking about what to put on those bays.

Pros I see:
- Cheaper per Tb
- Greater extensibility per chassis (3 times max capacity using 1.5 vs. 0.5)
It's the cons I'm wondering about. I expect using big disks to be slower, but is this really the case? I don't know about failure rates between small and big drives, are big ones expected to fail more often?

So, two questions:
1. What are the cons of building a 5Tb RAID5 with big disks?
2. Any particular drive model you'd recommend for either big (2Tb) or small (0.5Tb)?
- joeqwerty over 13 years
  
  Bigger drives != slower performance. The drive platters aren't physically bigger, they have more platters. Each surface of each platter normally has it's own read\write head (if each surface of the platter is used for data storage) so access times should be roughly equivalent to smaller drives.
- Steven Monday over 13 years
  
  On the contrary, I would expect bigger drives to be faster than smaller drives. All else being equal (rotational rate, number of heads, physical size), a larger drive will have a greater bit density than a smaller drive, which implies that its read/write rate will be higher.
- Chopper3 over 13 years
  
  It depends what you mean by 'big' and 'faster' - at any point in history the largest available capacity drives have always been pretty slow compared to the drives built for high load or outright performance (see today's 2TB 7.2krpm disk vs. 600GB 15krpm disks for example). Certainly bit density can help but so does rotational velocity, cache, interface speed, request latency, queuing capacity and a number of other factors. Either way rebuilding R5 arrays with consumer 1/2TB disks takes AGES and slaughters the drives during that time.
- Zan Lynx over 13 years
  
  Seeks are often slower on the big commercial drives because it takes longer to find the right data track.
Spence over 13 years

+1 - The bit error rate of disks makes errors a virtual certainty with large disks, as opposed to a theoretical risk. Very large RAID-5 sets carry with them the risk that you'll have a bit error during reconstruction. RAID 1+0 is the way to go for performance and availability.
Vinko Vrsalovic over 13 years

@Evan: That's all well and cool, but the cost is too high, especially when (as is my case) performance is non critical. So I guess RAID 6 it should be
Chopper3 over 13 years

what's the cost of losing the lot?
Vinko Vrsalovic over 13 years

Are the disks that advertise "works 24/7" really trustworthy? For example seagate.com/www/en-us/products/consumer_electronics/sv35_ser‌ies
Vinko Vrsalovic over 13 years

@Chopper3 That's a fine argument. Let's see if I can make it fly.
Leo over 13 years

@Vinko, No. the million plus MTBF is nonsense. (and IMO seagate is the worst.) That said, you "cannot not" buy them. How do you explain when a drive fails that you bought the desktop grade. So you are stuck buying paying extra for enterprise drives.
Chopper3 over 13 years

REALLY, REALLY NOT - those drives are for PVRs like Tivos etc. (in a bizarre coincidence I happen to buy literally hundreds of thousands of those disks, but I digress) - they'll SPIN fine for 24 hours a day, try thrashing it all day long and see it's MTBF go through the floor - I'm thinking more like these; seagate.com/www/en-us/products/enterprise-hard-drives/…
Steven Monday over 13 years

RAID != backup. You will have backups, right?
Vinko Vrsalovic over 13 years

Is a 500GB enterprise class HD safe enough for RAID5? (that's the other alternative for me)
Chopper3 over 13 years

@Vinko, it's as good as you can get but most, if not all, sysadmins on this site generally hate R5