So I assume the answer will be, its designed that way, let Unraid do it's thing, but here goes.

 

Given the fact that Unraid splits at a logical level (folder) and not at a size (block) there is no way to balance performance across disks. Sometimes you get lucky that the files are on several disks but you generally speaking you are stuck to the performance of a size drive for throughput. 

 

Even with "Most Free" which balances the data a bit better than "HighWater" data is being written file by file to only one drive.

 

If the answer is Cache, then yes I have one, right now a single disk (which I am adding more).  It definitely improves performance but is still locked to a single drive.(even when I add more)

 

Questions:

 

1. Am I correct in saying performance is mostly limited to a single drive? Am I missing something?

2. Data location decisions are based on file/directory not size, correct?

3. Do cache Raid configs follow the same rule and are file not size based?

4. Is a cache the only option to improve throughput?

 

 

Most Free can actually be slower for writing since parity has to update for each drive being written so no advantage to getting multiple disks writing at once. 

 

Do you know about Turbo write?

 

https://wiki.unraid.net/Manual/Storage_Management#Performance

 

Are you mostly concerned about read or write performance?

I did see Turbo write but got concerned when I read the issue concerning a "pause" while drives spin up. That would be bad. 

 

As for Read vs Writes, it depends on the share. Most of the array is general purpose NAS so it has about a 70/30 split with more reads. But I use SyncThing and that is a lot of random IO especially where the DB lives.

 

I guess I am just used to a typical NAS devise that uses 4-16K blocks and all drives spin all the time. You get to leverage the throughput of all drives which yields way better performance. My system is way overkill with multiple 10GB connections. I only see about 150MB/s max right now and that's only in spurts.

 

 

1 hour ago, DataDawg said:

I did see Turbo write but got concerned when I read the issue concerning a "pause" while drives spin up. That would be bad. 

Well you don't need to spin them down  There's also a plugin to automatically turn turbo on/off depensing on the number of already-running drives.

 

1 hour ago, DataDawg said:

But I use SyncThing and that is a lot of random IO especially where the DB lives.

The typical usage for apps on unraid is to have the application data (which would contain the db) on a a pool SSD, and the user data on the array.

I'll look into Turbo write since it seems to make sense. 

 

I do have my appdata on SSD and user data on SATA. I tried the "Yes" option for data but the 500GB SSD pool just kept filling and then performance really went down. I would need a much larger SSD pool to keep up with workload.

 

And I benchmarked "Most Free" vs "Highwater" and I found that performance was very similar except for large sequential writes which aren't common on my box. The benefit of having the it spread out for more even wear on the drives is worth the penalty at least in my use case.