NAS104 Raid decision? Xraid vs Raid 10 etc #24296869

offbyone wrote:
Write Speed: (io operations/ per write): Raid 5: 4x Raid 6: 6x Raid 10: 2x

I'm not sure how meaningful this analysis is, since it doesn't take caching into account

Disk caching (including deferred writes) is a very important piece of the puzzle, since if you are writing a sequential file you are writing all the data blocks in the stripe anyway, and that enables optimizations. If you make those optimizations, the combined i/o for those blocks is 1.33x in the RAID-5 case (you write 1 parity block for each 3 datablocks) and 2x for the RAID-6 case (you write 2 parity blocks for each 2 datablocks). For RAID-10, you are also writing 2x (one mirrored block for each data block). Also, the write queuing speeds things up in many practical use cases.

offbyone wrote:
Read Speed benefits: Raid 5: 3x Raid 6: 2x Raid 10: 4x

I am guessing here, but I think your benefits on the reads are not about the number of i/o operations, but they are about the ability to read the disks in parallel.
If so, the RAID 10 numbers are high. You can read both of the underlying RAID-0 volumes in parallel, so I am thinking 2x.

If you are talking about i/o generally, then Data blocks are spread across all four disks in all cases. RAID-5 requires you either to read parity blocks you don't need or seek over them. If you read them, then you are making 4 i/o requests for 3 blocks you care about, so like the write case the efficiency is 1.33x. Similarly RAID-6 has efficiency 2.0x, since you are reading 4 i/o requests and only getting 2 data blocks. RAID-10 does not match the write case, the efficiency there is 1.0x since every block is a block you care about. Of course if you seek, what happens depends on the seek performance of the drives.

Again, disk caching (including read-ahead) are very important aspects of the puzzle for sequential file access

offbyone wrote:

You point out that the 104 isn't really that fast. So does that normalize the results meaning that even though Raid 10 has great read/write speed benefits, you won't really see them? your point about the network speeds is not one I considered. My previous experience with raid was based as much around the speed benefits as the fault tolerance. But those were all environments where the raid array was connected directly or using a gigabit ethernet. Let's be honest, I surely plan to use the device via my wireless network where i live with speeds closer to 100mbs. I am not sure the effect though. Maybe it marginalizes the reads, but does make any write penalties inconsequential?

I think we can make this a bit simpler (excluding RAID-6). For large file sequential i/o:

The RN104 sustained write speeds (not including RAID-6) are measured at 320 to 400 mbits per second (40-50 MB/s). Sustained reads are similarly ~ 640 mbits per second.

On a wireless or fast ethernet network, the network is the bottleneck. So you will see close to 100 mbit performance on fast ethernet, and close to the real-world wifi speed for wireless. The NAS is fast enough that it doesn't get in the way.

When you have a faster network, then the RN104's processor performance becomes the bottleneck, and you will see 320-400 mbits of write performance or ~640 mbits per second of read performance no matter what RAID mode you pick (and no matter how fast the network is).

If you are running an application on the RN104 or doing random i/o, then of course it gets more complicated - and disk performance can become the bottleneck. And with a higher-end NAS, the disk i/o performance can become the bottleneck even with large sequential file access.

offbyone wrote:
that makes me lean towards raid 6 or raid 5.

I'd benchmark RAID-6 before I used it. Dual RAID-1 isn't quite as good protection, but I think it will be quite a bit faster. Doing all the parity block computations in software will slow the RN104 down. The simplicity of dual RAID-1's storage layout is attractive as well.

offbyone wrote:
The extra space sure would be nice, but I really want fault protection. I have seen a raid 5 array fail and become unrecoverable when a second drive failed during recovery.

Yes, that can happen. Running the scheduled maintenance can help, especially if you have a lot of files that aren't accessed much (which I have). Often the drive fails long before it is detected, because the bad patch simply hasn't been accessed in a while. I suggest running scrubs and disk tests quarterly, with defrags also quarterly. That amounts to one diagnostic test per month.

Of course this also points out the need for backup on another device.

offbyone wrote:
This is a tough decision. I wish I had more data. I may do some benchmarks. I wish netgear had a more complete one done.

Benchmarks on 6.2 would be good information to share if you have the time.