6.9.0/6.9.1 Introduced NIC Problem? No issues on 6.8.3. Should I downgrade?


Recommended Posts

I'm a bit in doubt about what to do.


Before the last two stable updates 6.9.0 and 6.9.1 I ran Unraid 6.8.3 without any problems with an Intel Quad Core i3-8100 on a Fujitsu D3644-B motherboard.

The one thing that was bothering me a bit was the write amplification when writing to my raid 0 cache pool of two Samsung EVO 970's.
Thankfully Unraid 6.9.0 seemed to have fixed this (after repartitioning the cache pool drives).

 

But now with 6.9.0 and 6.9.1 I run into a real stability issue with the network interface (so it seems) which leads to two things I didn't encounter before:

  • Transfer speeds dropping after a while.
    Copying files to a share over SMB from a remote Windows machine starts out at ~900mbit, but after a few gigabytes the transfer speed drops to a few megabits. After restarting the transfer the speed is back to normal, but drops again after a few minutes. I do not have this when copying from other machines like an Apple Macbook. But I also didn't have this problem before using Unraid 6.8.3 and transferring files from the remote Windows machine.
  • The NIC seems to trigger a kernel panic every now and then.
    This freezes network connectivity for a few seconds and then everything is back to normal. This mainly occurs when using a Windows 7 VM on Unraid itself and doing some heavy cpu load in that VM. Also didn't have this before with 6.8.3.

 

These two problems are introduced since upgrading to 6.9.0. To get rid of the problems I can downgrade, but when I downgrade back to 6.8.3 I have to re-partition my cache drives again as 6.8.3 doesn't recognise the new '1 MiB Partition Alignment'. This is not really a problem except for the write amplification bug that will be back. But this is less of an annoyance than the problems described above.

 

Below I have attached the part of the syslog with the NIC error stuff.
After Googling around a bit I tried setting 'TCP segmentation offload' and 'generic segmentation offload' to off (ethtool -K eth0 tso off && ethtool -K eth0 gso off) but it doesn't seem to help.

 

What should I do? 
Downgrade to 6.8.3 and accept the excessive writes on the cache ssd's or is there perhaps another thing I overlooked?

There is a bios firmware update available for my D3644-B. Perhaps I can try that, but then again using the current bios I had no problems on 6.8.3.

 

Thanks in advance for any advice/thoughts.

syslog.txt

Link to comment
On 3/15/2021 at 12:06 PM, Dennis84 said:

I'm a bit in doubt about what to do.


Before the last two stable updates 6.9.0 and 6.9.1 I ran Unraid 6.8.3 without any problems with an Intel Quad Core i3-8100 on a Fujitsu D3644-B motherboard.

The one thing that was bothering me a bit was the write amplification when writing to my raid 0 cache pool of two Samsung EVO 970's.
Thankfully Unraid 6.9.0 seemed to have fixed this (after repartitioning the cache pool drives).

 

But now with 6.9.0 and 6.9.1 I run into a real stability issue with the network interface (so it seems) which leads to two things I didn't encounter before:

  • Transfer speeds dropping after a while.
    Copying files to a share over SMB from a remote Windows machine starts out at ~900mbit, but after a few gigabytes the transfer speed drops to a few megabits. After restarting the transfer the speed is back to normal, but drops again after a few minutes. I do not have this when copying from other machines like an Apple Macbook. But I also didn't have this problem before using Unraid 6.8.3 and transferring files from the remote Windows machine.
  • The NIC seems to trigger a kernel panic every now and then.
    This freezes network connectivity for a few seconds and then everything is back to normal. This mainly occurs when using a Windows 7 VM on Unraid itself and doing some heavy cpu load in that VM. Also didn't have this before with 6.8.3.

 

These two problems are introduced since upgrading to 6.9.0. To get rid of the problems I can downgrade, but when I downgrade back to 6.8.3 I have to re-partition my cache drives again as 6.8.3 doesn't recognise the new '1 MiB Partition Alignment'. This is not really a problem except for the write amplification bug that will be back. But this is less of an annoyance than the problems described above.

 

Below I have attached the part of the syslog with the NIC error stuff.
After Googling around a bit I tried setting 'TCP segmentation offload' and 'generic segmentation offload' to off (ethtool -K eth0 tso off && ethtool -K eth0 gso off) but it doesn't seem to help.

 

What should I do? 
Downgrade to 6.8.3 and accept the excessive writes on the cache ssd's or is there perhaps another thing I overlooked?

There is a bios firmware update available for my D3644-B. Perhaps I can try that, but then again using the current bios I had no problems on 6.8.3.

 

Thanks in advance for any advice/thoughts.

syslog.txt 10.3 kB · 0 downloads

you might want to try:

disabling enhanced C1 power saving state (C1E) in the BIOS settings, or

disabling TCP checksum offloading with ethtool -K eth0 tx off rx off
disabling TCP Segmentation Offload (TSO) with ethtool -K eth0 tso off

 

try one at a time

Link to comment
  • 3 weeks later...

I'm new here (and new to actively using support forums in general), but i have similar issue, i think.

I'm using the trial license for a month now, started with version 6.9.0 on a relatively old and low performance system running a G3240 Pentium processor and 8GiB DDR3.

 

I'm using the system mainly as a media server using 3 2TB WD Red drives (1 parity) and no cache. as part of my evaluation process i've tested the transfer speech of the SMB shares and compared to my old WHS V1 machine that i intend to replace. I was getting relatively low transfer speeds that went even lower after a few GB of the file were transferred and CPU load was reaching 100%, after searching the forums i've changed the "Tunable (md_write_method):" to "reconstruct write" which helped a little but still i will get the high cpu load and the drop in transfer speed after a few GB.

 

reading further in the forum i found a post that suggested rolling back to 6.8.3, i did, and have a steady ~90-110 MB/sec transfer speed most of the time and CPU load is lower too.

 

I've installed Fix Common Problems plugin which warns me now that i should update to the latest version, which is probably a good idea but i don't want to lose the smb performance.

 

As I'm evaluating still and jumping back and forth between builds is pretty painless (so far) i think i'll upgrade and test the speed again. anyone here familiar with the issue?

 

i apologize in advance for maybe stealing this thread  :)

 

Link to comment
10 hours ago, gmadcow said:

any downside for staying on 6.8.3?

 

Not that I know of...

Well, maybe that you are missing out on some security updates coming with the newer Linux version used by 6.9.0+.

 

But I don't care about that because Unraid has been running stable as a rock since downgrading. No more NIC errors and proper SMB transfer speeds.

 

Even my 6.9.0 repartitioned cache drives still work in 6.8.3. I repartitioned them in 6.9.0 with the '1 MiB Partition Alignment' due to the write amplification bug on Samsung Evo's. But 6.8.3 doesn't complain about that. It only shows 'unknown' as the partition format, but everything works.

Edited by Dennis84
Link to comment
  • 5 months later...

Transfer still faster on 6.8.3 than on 6.9.2, CPU load is lower on 6.8.3.

I was planning on using an old d510 atom board which was under performing using 6.9.2 but ended up using a Pentium G3240 with higher CPU clock and score which helps.

 

What helped the most is setting 'Turnable (md_write_method)' in 'Disk Settings' to 'reconstruct write' instead of 'auto' or 'read/modify/write'. writing is faster but it involve reading from all the array drives in order to keep the parity data valid.

 

Still hope there is a way to lower the cpu load on SMB writes and it will be fixed/improved at some point...

Currently using 1Gb LAN and spinning HDDs so the ~114MB/sec write is fine (and only 4 drives in the array so the reconstruct write is not that big of an issue.

Link to comment

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Restore formatting

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.