0

One customer has got a Dell server with 16 x 1TB 7.2k Near Line SAS 2.5 HDD and the server will be configured according to my opinion. I cannot change specs as the order is already given.

I want to have 2 different hard drive partitions. I need the data to be recoverable while getting the most speed possible. I am normally a software developer but I need to tell the IT guys to make a specific RAID configuration.

My plan is to have a RAID6 configuration per 8 disks, so I will have 6TBx2 storage with a good two disk failure recovery. But I don't know if this setup would have downsides about speed or any other thing like recovery time after failure etc. Also my opinion is not based on previous experience but some tutorials/forum posts online.

8
  • 1
    First of all, note that 2 different hard drive partitions does not requires two RAID arrays. One drive with two partitions will do just fine. So you are not limited by this. Secondly: It depends on your goals. Read speed? Write speed? Is random access important? Is there one part which is important (e.g. where you stored a heavily used database) and is an other volume just for backups of logs? Think of all these things before deciding on a setup. Or even better. Think about, write them down and then discuss them.
    – Hennes
    Nov 5, 2015 at 11:21
  • A database and a file server will be running on those partitions. One in each. Also I know I can connect 16 disks in one array and get the same result, but wont the recovery take longer time then? A similar system with 2x4TB RAID0 serves enough now, i cannot forecast any bottlenecks.
    – Seyf
    Nov 5, 2015 at 11:28
  • You can measure how the current system is used. E.g. if the fileserver might mostly be reads. The DB might also mostly be reads, or it might do a lot of writing. Measure that. Then plan on how to set up. (Which BTW does not need to be symmetric. You could have 8 disks for the DB in four RAID1E's stripes to get huge read speed for the DB, and a RAID 6 for the mostly sequential fileserver reads. - So first measure, then plan, then discuss.
    – Hennes
    Nov 5, 2015 at 11:38
  • Everything is mostly read. The current server, as I said utilizes 10-50% of the RAID0 (no security at except for daily manual backups to another spare hdd). Does RAID1E support multiple disk failure? The company is non-technical. If anything they will find me to fix it probably.
    – Seyf
    Nov 5, 2015 at 12:04
  • What kind of RAID controller? What are your storage requirements (how much DB, how much mail). What are the availability requirements? Is there a backup plan? Because RAID is not a backup - RAID is for data availability, not data integrity (all the RAID in the world won't save you from rm -f -r *...) After all that, use a power-of-two number of data disks for any RAID-5/6 arrays, and adjust the segment/stripe size such that your fundamental write (file system or database block size) will match the size needed to write across all data disks and no more. Then align your partitions. Nov 5, 2015 at 12:43

1 Answer 1

2

It depends on your load, read/write ratio stats, supported modes in raid controller and so on. If you have HW RAID controller (you definitelly should), look for modes what supports. When I was building raids in simmilar configuration, I've built RAID50 - 4 arrays in RAID5 with 4 disks and one RAID0 across those arrays (it was primary fileserver with big files). It is fast and you can lost 4 "right" disks (one in each RAID5 array). Your supposed 2x RAID6 configuration will works too, just think about RAID0 over those RAID6 arrays.

And golden rule in disk array configuration is "Test it, test it, test it!". You can build one configuration, make some tests, destroy it, build another configuration and run tests again.

3
  • +1 for testing. (preferably with real production loads)
    – Hennes
    Nov 5, 2015 at 12:50
  • It is a compromise between speed, reliability, and capacity. Using RAID50 will give the capacity of 12 disks, but with two dead disks there will be 20% risk of data loss. RAID10 will only give the capacity of 8 disks, but it will be faster and only has 6.7% risk of data loss in case of two lost disks. RAID6 with two parity and one hot spare will give the capacity of 13 disks and have 0% risk of data loss in case of two lost disks. So in that respect RAID6 has better capacity and reliability, but it has worse write performance.
    – kasperd
    Nov 5, 2015 at 17:31
  • 1
    Yep, RAID6 is very good option for data reliability, but not each RAID controller (even good one) supports it. Nov 5, 2015 at 17:37

Not the answer you're looking for? Browse other questions tagged .