8

I am upgrading a server which has four drive bays. I'll be putting a new Adaptec RAID card into the server so it will support large drive sizes. I need to squeeze as much capacity as possible out of this four-drive cage, so 8 TB drives are my most likely choice.

This is a generic file server (but it handles recording video streams from a security system). I would like to use one of the drive bays for a hot standby, which leaves me with three disks to form the RAID array.

Although I could simply use two disks for mirroring + one hot spare, I've seen RAID 1E (spread over three disks + one hot spare, or RAID 5 three disks + one spare.

Is there a good reason to go one way or the other?

I read the question R5 vs R1E, but it doesn't really answer if R5 or R1E is better in the case of a good controller and multiple write streams.

Community
  • 1
TSG
  • 1,634
  • 6
  • 29
  • 51

4 Answers4

14

The answer hasn't been RAID 5 for at least a decade, it's as dead as punch-tape - use 4 disks and RAID 10 it.

Chopper3
  • 100,240
  • 9
  • 106
  • 238
  • 2
    But with R10 I no longer have a spare drive for rebuilding in case of failure. That's why I thought about R1E + spare. But what about 5E and 5EE ? They seem to be popular – TSG Feb 01 '16 at 17:07
  • 2
    The problem with cold sparing on giant drives is it takes so long and strains the disks so hard that double failures, once super rare, have become common. You're better off keeping the data in sync via RAID-10 and when you lose a drive, quickly replace replace it. – Steve Bonds Feb 01 '16 at 17:23
  • 6
    If you can afford the performance tradeoff, RAID6 on 4 drives is equal in capacity to RAID10, and provides better reliability. – David Feb 01 '16 at 18:53
9

You should carefully consider whether RAID 5 will work for you especially in a RAID rebuild situation. The amount of reads required to rebuild a RAID 5 array is significant and can fall foul of an Unrecoverable Read Error (URE) which will cause you to loose the array and have to recover the data from backups. For example typical 8TB drives have 1 URE per 10^14 bits read so for your array there is a significant chance (~85%*) that you will not complete the rebuild.

Do you and your data a favor, use RAID 10 and a hot swapable disk system. Keep a spare disk to hand.

*Yes that's a fairly naive view and you may get lucky but ... Did he fire six shots or only five?

user9517
  • 114,104
  • 20
  • 206
  • 289
  • Ok - so I'll skip R5. What about R1E (3 disks) +spare vs R1E (4 disks) vs R10 (4 disks) ? – TSG Feb 01 '16 at 18:06
  • https://www.google.co.uk/search?q=raid+1e+vs+raid+5&oq=raid+1e+v+raid – user9517 Feb 01 '16 at 18:13
  • 15% chance of completing the rebuild? Or *not* completing it? – user253751 Feb 02 '16 at 03:45
  • 3
    Which of those two numbers would be acceptable?!? – MadHatter Feb 02 '16 at 07:44
  • @immibis He said "not". So, 85% chance failure. (That is as of the answer's third version, after updates. This was flipped in an update after your comment was made.) – TOOGAM Feb 02 '16 at 19:39
  • @TOOGAMI flipped it in response to the comment as whilst it read fine to me I could see how people might miss understand it. – user9517 Feb 02 '16 at 19:41
  • 1
    @TOOGAM It previously said "a significant chance (<15%) of completing the rebuild". Which pretty clearly means that the rebuild will work only one time out of six (at best). In fact, what it says now *still* means that. (85% chance of not completing the rebuild = 15% chance of completing it) – user253751 Feb 02 '16 at 20:10
0

Storage admin here. I've build anything from simple RAID in a box to 144-drive infiniband ScaleIO. From what I read - I think, your answer is NOT a single box with a RAID in it.

As others pointed out,

  • Rebuild times are horrible no matter what you do in RAID level,
  • Chances of a successful rebuild will never depart the realm of "maybe".
  • It does not protect, at all, against
  • cage failure,
  • controller (RAID card) failure
  • power supply failure
  • OS issues requiring reinstall

So, for the money you will throw at a RAID card and multiple good 8TB drives you have on your mind, would you be better off if you opted for:

  • any cloud solution (Wasabi comes to mind)
  • Two servers instead of one, with some sort of mirroring solution (keeping it super simple for each box)
  • One super simple storage server with cloud backup

Also keep in mind, non-raid single disk has much less work to do in its lifetime, lengthening its life span. Also, it does not suffer from heat/vibration of its neighbors at all. So if I have to safeguard large data, frankly, RAID is not the way to go in the grand scheme of things. It used to be THE answer in 90's and maybe 2000's but definitely it's not the only game in town.

My recommendation:

  • Build a simple server. Boot from NVMe, no RAID gimmick, just use hardened PCIe like Intel P3700 series and the like. Boot OS from it. Use a singleton 8TB HDD, good one like like the HGST. MAYBE a two-drive RAID1.
  • Build another one and make it a replication partner one way or the other.
  • If one fails, HDD or otherwise, simply decommission it, make the partner primary, while you get the broken box on the bench and take your sweet time rebuilding it. When done, make it the mirroring partner.
Jun Sato
  • 101
0

Rebuild time with 8 TB will be horrendous. I've tested rebuild time of RAID-6 arrays of 8 HGST He8 drives on Adaptec 7xx5 controllers at more than 48 hours.

RAID-5 has been dead for many years. R1E is probably much faster at rebuilds, but will face the same danger of URE when rebuilding. With very good disks such as He8, I'd risk 1E; with any other model, I'd go with the (slow as hell) RAID-6, for safety.

wazoox
  • 6,782
  • 4
  • 30
  • 62
  • The problem is that I have for bays available. Raid 6 would require 2 data drives, 2 checksum drives. That's 50% loss...I wonder if that's the best use of space – TSG Feb 05 '16 at 03:14
  • @Telium well you have to choose between losing space and risking loss of data :) R1E is probably acceptable in terms of risk with most drives EXCEPT Barracuda. Sorry to name names, but really Seagate Barracudas are so unreliable that it's almost a joke. I have a pile of 130 failed drives next to my desk at work... 120 of them are Seagate Barracudas. – wazoox Feb 06 '16 at 10:20
  • I think this is boiling down to the 3 or 4 drive bay issue. If I spread the array over 8 drive I wouldn't care, but with only 2 data disks and 2 CRC disks I'm not sure R6 is right. Leaning towards 3 disk R1E + spare. – TSG Feb 06 '16 at 14:40