5

I'm new to system administration and I'm not really sure if this is possible. Our server supports up to 4 x 2 TB drives. We need a drive larger than 2TB for a particular use. We also need redundancy in the case of hard drive failure. We thought, maybe we could mirror 2 sets of the drives in hardware, giving us 2 logical drives in the operating system for a total of 4TB of space across the drives. Once we installed Windows Server 2012 we were hoping to be able to stripe them in the OS to give us the large drive we need. Maybe this is not possible. But what if we created an OS partition on one of the drives and created a smaller stripe? Is anything like this possible? What else would we be able to do?

Any confirmation or alternative solutions for the needs described would be greatly appreciated.

Adam
  • 87
  • 1
  • 1
  • 4

1 Answers1

-3

You can build up a hardware Raid 5 with 4 disks.

That will give you a drive with 6 TB. Do whatever you like with that drive.

And no - that Raid 5 will not normally be slower than a Raid 10 with 2 x 2 disks, because you will have 3 instead of 2 concurrent drives doing read/write-work. Hardware controllers nowaday can handle the type of load involved with raid5 checksummung without performance impact.

Nils
  • 7,695
  • 3
  • 34
  • 73
  • 6
    Nils - we try really hard around here to not recommend R5 any more due to the maths behind the whole 'guaranteed unrecoverable error on rebuild' thing - preferring R1/10 or R6/10 instead. – Chopper3 May 15 '15 at 13:03
  • 1
    @Chopper3 agreed with 6 disks or more. Up to 8 disks I prefer R6, over 10 disks can give R10 a performance bonus (if the controller is not the bottleneck), but has less reduncy than R6. So generally our currenct policy is R6. But not for this case here. – Nils May 15 '15 at 13:10
  • Keep in mind that R5/6 has parity whereas R1/0/10 does not. – Konrad Gajewski May 15 '15 at 13:25
  • 3
    @KonradGajewski are you suggesting R5/6 is more resilient than R1/0/10? Or are you just stating facts? Some of us can't tell from your comment. – MDMoore313 May 15 '15 at 13:50
  • 3
    RAID5 performance is indeed lower than RAID10 for writes on any array of 4 or more disks. You generally have 1 spindle worth of write performance in a given R5 array, with a slight boost for caching and/or sequential operations on a capable controller. – Hyppy May 15 '15 at 14:07
  • @BigHomie: It can be because of parity errors giving you warning in advance, before serious data starts to occur. Also it can detect errors that are not related to drives themselves. Personally I go for CRC in ReFS if I can. Here's an example: your RAID1 gives you errors. Which data (disk 0 or disk 1) do you consider valid? – Konrad Gajewski May 15 '15 at 14:33
  • 1
    @hyppy You are wrong. Measure the values with a current **hardware** RAID controller. – Nils May 15 '15 at 15:28
  • 3
    @Nils extraordinary claims require extraordinary evidence. 4 reads plus two writes sounds like a lot more than two writes to me. My systems agree. – Hyppy May 15 '15 at 15:53