Slow and Unresponisve Transfers Using SMB2

[TabascoJoe]

Hello
RT-AC68U with FW 382.2_beta3. USB3 EXT3 drive connected.
When SMB2 is enabled the transfer speeds going from the router to a Windows 10 system is slow. It starts out strong with a 20 to 30 MBs transfer, but quickly drops below 1 MBs. It will get to the point the transfer fails. This is on a wifi and wired connections.

Now when SMB2 is disabled the transfer speed is a solid 35MBs to 40MBs (btw WOW, best speeds out of any router I've seen).

I like to use SMB2 as SMB1 is not enabled on Windows 10 by default. I don't like the hassle of enabling it on all systems. What could be causing the erratic speeds on SMB2? Is it Windows or the AC68U? Thanks.

 

[dvohwinkel]

As I understand it SMB2 is much more compute intensive... but that wouldn't explain why it starts out at 20 and drops to 1

 

[TabascoJoe]

Why does the interface show Enable SMB2 protocol (default: No).
Is SMB2 that buggy that is not on by default?

 

[dvohwinkel]

Why does the interface show Enable SMB2 protocol (default: No).
Is SMB2 that buggy that is not on by default?

Probably because it is so much more compute intensive and is a relatively new addition to the firmware.

 

[RMerlin]

Why does the interface show Enable SMB2 protocol (default: No).
Is SMB2 that buggy that is not on by default?

Because it slows down performance by about 10% on older routers. It's only enabled by default on the RT-AC86U, as on that model it doesn't slow down performance in any measurable way.

 

[TabascoJoe]

With the FW 384.4 and newer the Changelog reads. "Ideally, people should change it to SMBv2 only". I went back to SMB2 and still the speeds are slow to unresponsive with the 2nd core max out at 100%. There is no way ideally I'm sticking to SMBv2 only. There has got to be an issue with the code as SMBv1 rock a solid 30 to 40MBS. I see above you state a 10% hit, I'm experiencing a 110% hit.

 

[TabascoJoe]

Just noticed that the max CPU and 0 bytes transfer is from the USB drive attached to the RT-AC86U transferring down to a Windows box (wired or wireless). If transferring up to the AC68U it's slow (around 5MBs) but never max CPU or 0 byte transfer.

 

[FTC]

Hi, did this problem ever got resolved ? I am facing the same problem and am looking for a bypass that still allows for smb2 to be used..

 

[Gedeon30]

Because it slows down performance by about 10% on older routers. It's only enabled by default on the RT-AC86U, as on that model it doesn't slow down performance in any measurable way.

Sorry to say this, but way more than 10%. In my experience speed is reduced about 80%. Probably due to some AES crypt or something similar not supported (or bad optimized) in ARM7 / 32bits ark.

 

[john9527]

Give this a try...
https://www.snbforums.com/threads/usb3-hdd-on-ac86u-with-merlin-fw.44271/page-2#post-402184

 

[RMerlin]

Sorry to say this, but way more than 10%. In my experience speed is reduced about 80%. Probably due to some AES crypt or something similar not supported (or bad optimized) in ARM7 / 32bits ark.

What does AES crypt have to do with Samba?

10% was the numbers I had throughout my tests while implementing this.

 

[Gedeon30]

What does AES crypt have to do with Samba?

10% was the numbers I had throughout my tests while implementing this.

It was just a guess since SMB v2 is safer but way slower (at least in the AC68/AC66_B1) I tend to think is due to some kind of crypt/algorithm which is bad supported by certain SoCs/compilers/kits.

Like those SSSE3 or AvX extensions which you can't find in chips which are more than 5-6 years old.When apps/compilers don't find certain extensions they choose another route to solve the maths, way more basic and compatible, but way slower too.

It's a guess from the behaviour I've seen. Granted SMBv2 is at least about 4-5 times slower in my router than SMBv1 (one large file through wifi). I personally don't have problems keeping SMB v1 in my home. I'm not running any critical service and compatibility is something I value.

 

[RMerlin]

It was just a guess since SMB v2 is safer but way slower (at least in the AC68/AC66_B1) I tend to think is due to some kind of crypt/algorithm which is bad supported by certain SoCs/compilers/kits.

Samba 3.6.xx doesn't use any encryption.

If your performance drop is higher than 10% then something else is affecting your performance, it's not caused by the SMBv2 protocol alone.

 

[FTC]

Hi, it is not a 'performance drop' per se, but a sporadic slowdown to as low as 0Kb/sec while transferring large files with SMB2.., as it has already been reported by a few users in the last couple of months (like for instance here https://www.snbforums.com/threads/smb2.42048/#post-356940).

 

[Gedeon30]

Samba 3.6.xx doesn't use any encryption.

If your performance drop is higher than 10% then something else is affecting your performance, it's not caused by the SMBv2 protocol alone.

I don't know the reason. Maybe it's just a bug, a buffer, no idea. The smbd process uses one core at 100% with the SMBDv2 in the AC66_B1.

I've just replicated the test again with the same results.

The first test I did with 384.4 and Windows 10 1709. Today I've done with 384.5 beta 2 and Windows 10 1803.

 

[RMerlin]

Hi, it is not a 'performance drop' per se, but a sporadic slowdown to as low as 0Kb/sec while transferring large files with SMB2.., as it has already been reported by a few users in the last couple of months (like for instance here https://www.snbforums.com/threads/smb2.42048/#post-356940).

There were some experiments a few months ago regarding the Samba config file, some optimizations done by these users showed good improvements in some scenarios (especially over wifi I believe). I suspect that playing with these settings might help.

 

[Adamm]

In this thread we tested various config options and made some significant improvements with SMB, on the AC86U we were able to improve read and write speeds around x2 - x3. Older models the results weren't as high but still much better then default configuration

 

[Qing]

Give this a try...
https://www.snbforums.com/threads/usb3-hdd-on-ac86u-with-merlin-fw.44271/page-2#post-402184

In this thread we tested various config options and made some significant improvements with SMB, on the AC86U we were able to improve read and write speeds around x2 - x3. Older models the results weren't as high but still much better then default configuration

I had the same problem, and it was perfectly resolved by setting "socket options = IPTOS_LOWDELAY TCP_NODELAY SO_KEEPALIVE".
Thank you very much.

 

[FTC]

I had the same problem, and it was perfectly resolved by setting "socket options = IPTOS_LOWDELAY TCP_NODELAY SO_KEEPALIVE".
Thank you very much.

Hi, thank you for the suggestion. Just tried it (I had missing the SO_KEEPALIVE thing) but unfortunately I am still experiencing the slowdowns when copying from the router's samba share through wifi. (does not happen when copying from another windows 10 machine share through wifi)

 

[Trebuin]

I did some tests on SMB 1 vs 2 on my Mac & it shows that SMB 1 has a much more stable (same speed) connection whereas SMB 2 tends to have a lot of up & down spikes with what should be a higher throughput. In the end, SMB 1 seems more decent for streaming purposes for me & the difference in overall long term speed is almost negligible. This is with the AC3200.