Slow Write Speed

[fluffybunnyuk]

I have a pair of 12TB drives in a  seperate pool as a pseudo mirror (0 file fragments). But am getting 20mb/sec write speeds, and 40mb/sec reads.

The kick in the teeth is duplication is 80mb/sec.

When i test the drives, i get 450mb/sec bus speed per port. When i copy a file to a single drive bypassing drivepool i get 250mb/sec. As soon as i use drivepool performance craters.

I've tried disabling real time duplication, disabling file balancers, but nothing seems to shift  the performance, it just gets slower and slower the more i add.

It started out at 100mb/sec, dropped to 80mb/sec, then 60mb/sec, now i'm down to 20mb/sec and getting slower.

Edited May 2, 2021 by fluffybunnyuk
forgot to add timesscale of performance

[fluffybunnyuk]

Solved: Moving and defragging the MFT and related $ files to a location 5% into the platter, and a reboot.I extended the MFT to a 1 GB(700mb free) in size just to be on the safe side.

Yay sustained 200mb+/sec is back.

[Christopher (Drashna)]

That's good information to know!

 

Though, would you be willing to post the process of how you did all this, specifically? 

[fluffybunnyuk]

Ultimate defrag has a boot time defrag for drive files like $MFT etc

I go into tools/settings and then click the boot time defrag option.

I set the starting cluster to 100000000 (12tb disk)(you can set your own location but 5% in seems reasonable for 0% to 10% track seek for writes, 10%  into the disk would be optimal(0-20% seek) .and leave the order as the default. I change the free space after the reserved zone to a reasonable size (in my case 752mb to round it to a 1GB MFT)

I click the run the next time the system starts box and click ok.

Naturally a warning needs to be attached to this. Your playing with your disks file tables.... It can go really bad (unlikely but possible if your system is unstable). So to mitigate that risk a chkdsk /F MUST be run on the drive BEFORE. And preferably the disk needs as little fragmentation as possible( preferably a complete fragmented file defrag pass first)..

 

As can be seen from the pic. Most of my data is moved to the rear , so all my writes happen at the start of the disk near the MFT, this means the head doesnt move too far when updating the MFT. Contiguous Reads i dont care about (i dont need speed other than pesky tape drive feeding), and writes are crucial to me to be executed asap since i dont use cache ssd.