Setting up a RAID 5 with expansion in the future in mind
I am setting up a NAS box, but money is a bit short right now. I am planning on getting the box up and running with a boot drive, then 3 for a RAID 5 config with some lower end drives.
I was reading online about a bunch of the complications that can arise from people not setting it up right, and couldn't get a feeling of exactly how hard/feasible it would be.
So,
- Is it possible to upgrade (replace a drive with a bigger/faster one) and/or add drives (without wiping, and rebuilding)?
- How risky is it to do this with data that isn't backed up (because of sheer volume)?
- Any tips/advice I should follow to set up a best case in the future to allow upgrades?
Note, I'm really between a RAID 5, and a RAID 01. Would this be easier to upgrade a RAID 01?
Instead of a single boot disk and 3 disks in RAID 5, consider 4 disks in RAID 10. This is both faster and safer. You do not want your NAS to go down if the single boot disk fails. Additionally, RAID 5 has poor write performance and does not rebuild reliably on large, terabyte-denominated volumes, especially when you are using inexpensive drives.
Answers:
Replacing a disk with a larger one: Not with a standard RAID controller. Some vendors (e.g. Drobo) have proprietary RAID implementations that support adding disks of dissimilar sizes, etc., but traditional RAID remains far more dependable. If you want to play with something dynamic and innovative, I would suggest ZFS. Added in response to TomTom's observation that only half of question #1 had been answered: Expanding an array by adding disks of the same size as existing array members is supported by many enterprise RAID controllers. Larger disks will be treated as though they are no larger than existing disks. This may be unsupported by low-end/integrated RAID devices. Check the documentation for your specific controller.
No. RAID is never an excuse for failing to have a backup and a DR plan. Having "too much data" is also not an excuse for failing to back up; however, it may be an opportunity to assess whether any significant portion of your data can be classified as genuinely unimportant.
With consumer grade 2TB SATA drives costing only $80 each in 2011, and enterprise-grade 2TB SAS drives costing only $240, you can add 4TB of usable RAID 10 capacity for $320 (consumer-grade) or $960 (enterprise-grade) if you have sufficient bays and connectors. That's a very cheap incremental upgrade path.
Re 1: depends. On technology used, which you say nothing about. I use Adaptec raid controllers, and thy do that, without pulling the raid.
Re 2: depends how much you value your data. I copy all data to another server or two with large slow drives.
Re 3: Except investing oney into a decent raid controller and making sure you have the hardware layout in place for that? For example, what is expansion for you? I run a 24 slow server for that reason (to allow expansion), rack units. Now i start filling up the last 8... the next server will be a 4 rack unit 72 bay machine allowing me to put in 72 discs. But then, this costs (just th case) around 3000 USD, so "cheap" and "expansion" are very relative terms, you know.
Regarding question 1:
Yes, it is feasible to expand a RAID 5 by either adding more disks or by replacing existing disks with larger disks. "The maximum number of drives in a RAID 5 redundancy group is theoretically unlimited." [1]. Ask yourself first, however, whether your filesystem will support this operation. Adding larger disks needs to be done across the entire RAID before you will be able to recognize the benefit.
Regarding question 2:
I would characterize a RAID reconstruction as inherently risky. I do it all the time but I never do it without careful thought and a bit of nervousness.
Regarding question 3:
Look at LVM. Also, consider using an external JBOD (just a bunch of disks) instead of trying to cram all the drives into a standard chassis. Disks are relatively cheap. Go SATA. You can get 3TB SATA drives for $150. Buy the best RAID card you can afford. I've had good luck with LSI and Adaptec.