Finally moved data from my old nas over to unraid. While balancing out data across drives, I noticed the turbo write speeds are much slower than when I was testing on different drives. I wonder if there is anything I can check to see whats wrong.

 

While testing with 3 laptop hard drives (1tb parity + 2x 500gb), I was getting around sustained 90-100 MB/s turbo write speeds while moving data around between drives with unbalance plugin. All the other system hardware was the same. This testing was with larger data files. It was writing between hard drives and not to/from cache disk.

 

On my main setup (3tb parity + 3tb + 2x 2tb drives), I see that the turbo write speed is always limited by the read speed of the drive that isn't storing the data being moved. As seen in the screenshot below, the drive holding the data is usually reading at high speeds, but the other drive reading parity bits often drops to low speeds (10-40MB/s). Sometimes the drive not storing the data can read at much higher speeds during a turbo write xfer, but it is only for short durations. This is with large data file transfers. I have reconstruct write on, NCQ on, and turned off dockers/vms and other things that might affect it like folder caching, but that didn't fix the issue. Typically for large folder transfers, the avg transfer speed is ~40MB/s with turbo write. These drives are faster than the laptop hard drives that I tested on. When doing a parity check, they all read at 100+ MB/s consistently.

 

 

 

 

Edited August 8, 20187 yr by kakashisensei

The 2nd pic show running parity check, so any write speed test would be slow.

I wasn't testing write speeds when the parity check was running. That was just to show there should be no reading speed problems on the array.

 

I did some more testing. When I write from cache to array, then the disks doing the reading get consistently high speeds. Its only when I move data from one disk to another disk in the array that there is much slower turbo write speeds.

Edited August 8, 20187 yr by kakashisensei

Note.

 

35 minutes ago, kakashisensei said:

When I write from cache to array, then the disks doing the reading get consistently high speeds.

That means array ( speed ) was normal.

 

35 minutes ago, kakashisensei said:

I move data from one disk to another disk in the array that there is much slower turbo write speeds.

I haven't deeply study above case, but some situation turbo-write slow was normal, that's why have 2 mode and @Squid make a plugin.

 

Edited August 8, 20187 yr by Benson

49 minutes ago, kakashisensei said:

When I write from cache to array, then the disks doing the reading get consistently high speeds. Its only when I move data from one disk to another disk in the array that there is much slower turbo write speeds.

 

If you copy or move files from one disk to another with turbo write enabled twice as many blocks have to be read from the source disk as the other disks: first the source file has to be read and then, after a seek operation, the same disk is read once again for the parity calculation. The destination disk is written once and so is parity. All the remaining disks are read once. So the source disk is working harder than the others.

 

7 hours ago, John_M said:

 

If you copy or move files from one disk to another with turbo write enabled twice as many blocks have to be read from the source disk as the other disks: first the source file has to be read and then, after a seek operation, the same disk is read once again for the parity calculation. The destination disk is written once and so is parity. All the remaining disks are read once. So the source disk is working harder than the others.

 

Thanks for this - I didn't realise.  I'll now turn off TW if I'm doing big Disk-2-Disk transfers.

Like John_M mentioned Turbo Write will not perform its best doing a disk to disk copy, but generally it's still a little faster than with it disable, depending on the hardware, so try both ways.

I should have posted this here instead: