December 30, 2016

Building a NAS Part 1

The Problem

I am sick of losing data. I really value the data I have and I simply do not want to lose it.

So I have decided to build a NAS to consolidate and most importantly replicate my data.

The Solution

At minimum, I want some kind of network-accessible storage array that can withstand one disk failure without loss in uptime and then back up that array locally and remotely.

This strategy addresses disk failure, array failure (electrical surge, theft), catastrophic disaster (fire).

My data needs are modest. I have less than 1 TB of data that needs to be backed up. This amount is unlikely to grow. I estimate that 2 TB is the most I will ever need to store. Accessibility of the data is not terribly important, so uptime is not critical.

I then decided one some kind of low powered computer that could support an array of 2-4 disks. I also decided to use the FreeNAS operating system.

I am still debating the array configuration and hard drive choice.

The Drives

  • ‘New’ 2TB enterprise drives with a one year seller warranty for approximately $50 each on Ebay. These drives are apparently new drives that did not sell. They are warranted by the seller, not the manufacturer. I have considered these drives because they are cheap, have some warranty, and my backup plan allows for unreliable drives. If the array fails, I have all my data locally backed up.

  • 3 TB Toshiba P300 with a 2 year manufacturer warranty. These drives are new, regular desktop drives commonly available for $80 each from places like Newegg.

I have not considered new NAS designed drives (eg WD Red) because of expense. A new 2TB WD Red costs $90 and a new 3 TB model runs $105. Their only advantage is NAS optimized firmware and a 3 year warranty. The problem is that these drives command at least a 25% price premium. I’m sure these drives are better choices, but I don’t think their advantages are worth the 25% premium in my situation. (Low usage, highly redundant, not important uptime)

The Array

I have considered a few options:

  • 3 drive mirror of ‘new’ 2TB disks. $150, 2 TB capacity, 2 disk failure tolerance

  • 3 drive mirror of 3 TB Toshiba disks. $240, 3 TB capacity, 2 disk failure tolerance

  • 4 drive Z2 array of ‘new’ 2 TB disks. $200, 4 TB capacity, 2 disk failure

I have eliminated a 2 drive mirror from consideration because I can achieve 2 disk failure tolerance for $150, which is in my budget and very much worth it. If I can afford it, there is no reason to save money to drop to 1 disk failure tolerance.

I have also eliminated a 4 drive stripe of (2 drive) mirrors. (Aka Raid 10) This array achieves the same capacity of a Raid Z2 array, but with the caveat that after a single disk dies, the other drive in the mirror cannot also fail. Either drive in the other mirror can fail, but not not the mirror’s partner. A Z2 array can withstand any second disk failure. Now the odds of the other mirrored disk failing is pretty small, but a Z2 array achieves better fault tolerance at no additional cost. I’m sure there’s some performance difference, but I really doubt I will see the difference between the two array types in my usage scenario.

Performance of the different arrays is something I do not fully understand. I am ultimately limited by gigabit ethernet. Also, this array will only be used for backup and storage purposes. IO performance is not critical. As long as I can get acceptable performance out of gigabit ethernet, I will be happy. I am willing to sacrifice a little performance for redundancy, but performance still needs to be acceptable.

The only benefit of a Z2 drive array seems to be capacity. Capacity above 2 TB is not hugely beneficial to me. Sequential performance seems to be improved in a Z2 array compared to a mirror, but hard drives can achieve near 100 MB/s performance and gigabit ethernet is limited to 125 MB/s. Gaining performance beyond a single hard rive won’t be hugely beneficial to me either.

Conclusion (for now)

I am leaning on a 3 drive mirror of the ‘new’ 2 TB enterprise drives. The drives are cheap and reasonably reliable since they are unused and have a 1 year seller warranty. If one fails outside the 1 year window, $50 is not a huge cost to pay. A 3 drive mirror seems like the best option that meets my reliability requirement while having the highest cost to performance ratio.