Hey guys, so i have recently added a 10gbe card aqc107 to my unraid server, the same chip is in my desktop.

 

I have a unraid array etc but thats not going to be relevant in this instance, i have a NVME cache drive infact the exact same drive in both client and server, a Samsung 960 evo 256gb.

 

When i am copying from a cache enabled share to my local pc C:\ (nvme) i get almost solid 1GB/s as shown

 

However if i copy the same file back to again a cache enabled share i get

 

Now this is too fast to be a HDD so it has to be the NVME but why is it only ~250 and not 1GB

 

I almost expect it to be the other way, where my writes are slower and not my reads?

 

Windows client SSD Benchmark 

Edited December 8, 20196 yr by Fma965

Try enabling direct i/o, if still the same see if reading directly from the disk share is faster (//tower/cache instead of //tower/share)

5 hours ago, johnnie.black said:

Try enabling direct i/o, if still the same see if reading directly from the disk share is faster (//tower/cache instead of //tower/share)

Thanks, I have tried with and without direct io, not really any noticeable difference.

 

I will try reading it from the cache disk share later, is there a reason that would make a difference? 

Edited December 8, 20196 yr by Fma965

I have just tried this and it unfortunately hasnt made any difference, i am using a cat5e cable currently (which although not officially specced for 10gbe does work) i have a cat6 cable coming tomorrow so will test with that.

Edited December 8, 20196 yr by Fma965

16 hours ago, Fma965 said:

I will try reading it from the cache disk share later, is there a reason that would make a difference? 

Disk shares can be faster, in some cases much faster, than user shares.

10 hours ago, johnnie.black said:

Disk shares can be faster, in some cases much faster, than user shares.

fair enough, unfortunately not in this case. i have just tried a cable that claims to be a cat6 cable and no change.

 

I also swapped the 2 cards over and the issue persists the same way, reading from the server is slow despite the cards being reversed now (indicating a software issue surely?)

First run iperf to confirm lan bandwidth then you can try copying to a ramdisk to rule out other bottlenecks, I can get 1GB/s reading from my cache pool.

 

 

But it's slower if writing to my desktop since the NVMe devices can't keep up:

21 minutes ago, johnnie.black said:

First run iperf to confirm lan bandwidth then you can try copying to a ramdisk to rule out other bottlenecks, I can get 1GB/s reading from my cache pool.

 

 

But it's slower if writing to my desktop since the NVMe devices can't keep up:

I thought i already mentioned this but it seems not, talking to so many people about this issue...

 

iperf has the same

 

When unraid is the server "-s" i get this

 

When my PC is the server "-s" i get this

 

I'm running out of ideas on what the issue could be.

 

Do you have jumboframes on or off? i've tried with both.

4 minutes ago, Fma965 said:

iperf has the same

OK, so the problem is network related, if iperf doesn't get more than 2Gbits, neither will any single transfer, you need to try different things, like NIC, another PC, etc until iperf performs normally.

17 minutes ago, johnnie.black said:

OK, so the problem is network related, if iperf doesn't get more than 2Gbits, neither will any single transfer, you need to try different things, like NIC, another PC, etc until iperf performs normally.

i dont' have any other 10gbit equipment so not able to do any other testing, the iperf works fine 1 direction but not the other. 

 

EDIT: i have figured it out i think, i just tried unraid to unraid via a test usb and it worked at full speed, i double checked my windows settings and saw MTU was on 9000 on Unraid but on windows the jumboframe setting had turned it self back off so i guess thats the issue.

Edited December 9, 20196 yr by Fma965

Yeah, all is working now, stupid windows reverted my setting.