I have a 500GB cache pool of 2 x 500GB SSDs

 

Transferring files to the SSD cache rarely goes over 80MB/s, and often drops to 20MB/s

 

The SSDs are connected to SATA2 ports onboard my Supermicro mainboard, but this should still be plenty to saturate gigabit

 

Even if i transfer files from within a Windows Server VM inside the cache pool, it still sticks at the above speed

 

I have the trim plugin running daily, but the performance i am getting suggests the drives are not trimmed

 

Any ideas?

Edited March 27, 20179 yr by sdamaged

Post your diagnostics:

 

Tools -> Diagnostics

 

 

Ok here it is, thanks

ba-unraid-diagnostics-20170327-1246.zip

Those are right after a reboot, post new ones after more uptime and at least one scheduled trim.

Ok here is another after a couple of trims

ba-unraid-diagnostics-20170327-1612.zip

Trim seems to be working correctly, I would suggest testing the pool write speed to rule out any network issues, download the attached script to the root of your flash share and run it like so:

 

/boot/write_speed_test.sh /mnt/cache/test.dat

 

write_speed_test.sh

writing 10240000000 bytes to: /mnt/cache/test.dat

1401092+0 records in
1401092+0 records out
1434718208 bytes (1.4 GB, 1.3 GiB) copied, 5.0004 s, 287 MB/s
2644339+0 records in
2644339+0 records out
2707803136 bytes (2.7 GB, 2.5 GiB) copied, 10.0037 s, 271 MB/s
2776247+0 records in
2776247+0 records out
2842876928 bytes (2.8 GB, 2.6 GiB) copied, 19.0049 s, 150 MB/s
3028005+0 records in
3028005+0 records out
3100677120 bytes (3.1 GB, 2.9 GiB) copied, 20.0124 s, 155 MB/s
4336051+0 records in
4336050+0 records out
4440115200 bytes (4.4 GB, 4.1 GiB) copied, 25.0157 s, 177 MB/s
5656775+0 records in
5656775+0 records out
5792537600 bytes (5.8 GB, 5.4 GiB) copied, 30.0191 s, 193 MB/s
6362486+0 records in
6362486+0 records out
6515185664 bytes (6.5 GB, 6.1 GiB) copied, 41.6155 s, 157 MB/s
7260401+0 records in
7260401+0 records out
7434650624 bytes (7.4 GB, 6.9 GiB) copied, 45.0297 s, 165 MB/s
8598296+0 records in
8598296+0 records out
8804655104 bytes (8.8 GB, 8.2 GiB) copied, 50.0329 s, 176 MB/s
9785584+0 records in
9785584+0 records out
10020438016 bytes (10 GB, 9.3 GiB) copied, 55.0365 s, 182 MB/s
9813716+0 records in
9813716+0 records out
10049245184 bytes (10 GB, 9.4 GiB) copied, 63.6835 s, 158 MB/s
10000000+0 records in
10000000+0 records out
10240000000 bytes (10 GB, 9.5 GiB) copied, 64.5865 s, 159 MB/s
write complete, syncing
removed '/mnt/cache/test.dat'

Edited March 27, 20179 yr by sdamaged

So it's not a cache pool problem, probably network.

Edited March 27, 20179 yr by johnnie.black

OK so i think i've ruled out network as best as i can

 

I reset the network to just use a single gigabit port, rather than the four it was using 

 

I copied an 80GB mkv file over the network to a share, and it averaged 80MB/s but with lots of long drops down to below 5MB/s and even hitting 800KB/s, ensuring the transfer took a long time, occasionally even stopping altogether

 

I kept an eye on network bandwidth using netdata and the network port wasn't at capacity, and my CPU wasn't being hammered either. Also checked the cache drives inside netdata and they are also not being maxed out (which you would expect as they can take 250MB/s on SATAII)

 

Also, as this was going to the cache drive, the two parity drives weren't being used (i had wondered if they could have been an issue as they are SMR Seagate Archive drives), but the parity obviously isn't calculated until it hits the main array.

 

I am completely baffled here