Bad TCP performance on WLE900VX

Bad TCP performance on WLE900VX

[Robert Jones]

Hello Everyone,

I'm using a custom arm64 board with a Cavium CN8030 CPU and testing
wireless card performance via iperf on Ubuntu 16.02 with a mainline
4.14 kernel. I have two identical units connected to each other via
60dB attenuators on all three channels. Channel is set to 161 VHT80.

With UDP I see throughput upwards of 650Mbits/s, but TCP is
~24Mbits/s. That seemed odd so I did a little more testing. Switched
to a 4.9 kernel and still saw the poor performance, but not with a
4.4. Before I went through the trouble of bisecting I thought to ask
if there was something I was missing.

I did come across this Ubuntu bug thread
[https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1670041] which
appeared to follow the same behavior I was seeing. Switched my TCP
congestion algorithm to reno and saw throughput jump to >330Mbits/s.
Cubic and BBR still ~24.

My question now, is there a standing bug that I'm not aware of or
could there be some sort of problem with my configuration that is
being worked around with the reno algorithm?

Thanks,
Robert Jones

_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

Re: Bad TCP performance on WLE900VX

[Ben Greear]

On 01/02/2018 05:47 PM, Robert Jones wrote:
> Hello Everyone,
>
> I'm using a custom arm64 board with a Cavium CN8030 CPU and testing
> wireless card performance via iperf on Ubuntu 16.02 with a mainline
> 4.14 kernel. I have two identical units connected to each other via
> 60dB attenuators on all three channels. Channel is set to 161 VHT80.
>
> With UDP I see throughput upwards of 650Mbits/s, but TCP is
> ~24Mbits/s. That seemed odd so I did a little more testing. Switched
> to a 4.9 kernel and still saw the poor performance, but not with a
> 4.4. Before I went through the trouble of bisecting I thought to ask
> if there was something I was missing.
>
> I did come across this Ubuntu bug thread
> [https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1670041] which
> appeared to follow the same behavior I was seeing. Switched my TCP
> congestion algorithm to reno and saw throughput jump to >330Mbits/s.
> Cubic and BBR still ~24.
>
> My question now, is there a standing bug that I'm not aware of or
> could there be some sort of problem with my configuration that is
> being worked around with the reno algorithm?

There are more tcp related hacks that make ath10k useful for
generating TCP traffic.  The tcp_tsq_limit_output_interval
one in particular is helpful.

You can dig through my kernel and look for a commit related to
tcp_tsq_limit_output_interval.  Try setting it to 5 or higher.

https://github.com/greearb/linux-ct-4.13

Thanks,
Ben


-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com

_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

Re: Bad TCP performance on WLE900VX

[Kalle Valo]

Robert Jones <rjones@gateworks.com> writes:

> Hello Everyone,
>
> I'm using a custom arm64 board with a Cavium CN8030 CPU and testing
> wireless card performance via iperf on Ubuntu 16.02 with a mainline
> 4.14 kernel. I have two identical units connected to each other via
> 60dB attenuators on all three channels. Channel is set to 161 VHT80.
>
> With UDP I see throughput upwards of 650Mbits/s, but TCP is
> ~24Mbits/s. That seemed odd so I did a little more testing. Switched
> to a 4.9 kernel and still saw the poor performance, but not with a
> 4.4. Before I went through the trouble of bisecting I thought to ask
> if there was something I was missing.
>
> I did come across this Ubuntu bug thread
> [https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1670041] which
> appeared to follow the same behavior I was seeing. Switched my TCP
> congestion algorithm to reno and saw throughput jump to >330Mbits/s.
> Cubic and BBR still ~24.
>
> My question now, is there a standing bug that I'm not aware of or
> could there be some sort of problem with my configuration that is
> being worked around with the reno algorithm?

There's now support in TCP stack (commit below) to workaround this issue
in ath10k. I tried to come up with a patch before holidays but didn't
make it, anyone willing to write one?

tcp: allow drivers to tweak TSQ logic
https://git.kernel.org/linus/3a9b76fd0db9f0d426533f96a68a62a58753a51e

-- 
Kalle Valo
_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

Re: Bad TCP performance on WLE900VX

[Robert Jones]

> There are more tcp related hacks that make ath10k useful for
> generating TCP traffic.  The tcp_tsq_limit_output_interval
> one in particular is helpful

> You can dig through my kernel and look for a commit related to
> tcp_tsq_limit_output_interval.  Try setting it to 5 or higher.

This solved the problem for me.

Link to commit for anyone else looking:
https://github.com/greearb/linux-ct-4.13/commit/dc201c3c310f805681dbf393484bdc9715745df1

Thanks!
Robert

_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

Re: Bad TCP performance on WLE900VX

[Ben Greear]

On 01/10/2018 10:15 AM, Robert Jones wrote:
>> There are more tcp related hacks that make ath10k useful for
>> generating TCP traffic.  The tcp_tsq_limit_output_interval
>> one in particular is helpful
>
>> You can dig through my kernel and look for a commit related to
>> tcp_tsq_limit_output_interval.  Try setting it to 5 or higher.
>
> This solved the problem for me.
>
> Link to commit for anyone else looking:
> https://github.com/greearb/linux-ct-4.13/commit/dc201c3c310f805681dbf393484bdc9715745df1

Out of curiosity, did you try the patch that Kalle suggested?

I'd be curious how that performed in comparison...

Thanks,
Ben


-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com


_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

Re: Bad TCP performance on WLE900VX

[Robert Jones]

> Out of curiosity, did you try the patch that Kalle suggested?

I did not unfortunately. I have other cards to qualify but if I can
come back to this I'll post the results.

On Thu, Jan 11, 2018 at 2:50 PM, Ben Greear <greearb@candelatech.com> wrote:
> On 01/10/2018 10:15 AM, Robert Jones wrote:
>>>
>>> There are more tcp related hacks that make ath10k useful for
>>> generating TCP traffic.  The tcp_tsq_limit_output_interval
>>> one in particular is helpful
>>
>>
>>> You can dig through my kernel and look for a commit related to
>>> tcp_tsq_limit_output_interval.  Try setting it to 5 or higher.
>>
>>
>> This solved the problem for me.
>>
>> Link to commit for anyone else looking:
>>
>> https://github.com/greearb/linux-ct-4.13/commit/dc201c3c310f805681dbf393484bdc9715745df1
>
>
> Out of curiosity, did you try the patch that Kalle suggested?
>
> I'd be curious how that performed in comparison...
>
>
> Thanks,
> Ben
>
>
> --
> Ben Greear <greearb@candelatech.com>
> Candela Technologies Inc  http://www.candelatech.com
>



-- 
Robert Jones - Software Engineer
Gateworks Corporation - http://www.gateworks.com/
3026 S. Higuera St. San Luis Obispo CA 93401
805-781-2000

_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

Re: Bad TCP performance on WLE900VX

[Sebastian Gottschall]

Am 12.01.2018 um 02:24 schrieb Robert Jones:
>> Out of curiosity, did you try the patch that Kalle suggested?
> I did not unfortunately. I have other cards to qualify but if I can
> come back to this I'll post the results.
this patch wont change anything, since you still need to modify ath10k 
to set a sane tsq value
see this additional patch from lede
https://git.lede-project.org/?p=openwrt/openwrt.git;a=commitdiff;h=111b49902465116a8353d29afe02eff0f56ea0a3
>
> On Thu, Jan 11, 2018 at 2:50 PM, Ben Greear <greearb@candelatech.com> wrote:
>> On 01/10/2018 10:15 AM, Robert Jones wrote:
>>>> There are more tcp related hacks that make ath10k useful for
>>>> generating TCP traffic.  The tcp_tsq_limit_output_interval
>>>> one in particular is helpful
>>>
>>>> You can dig through my kernel and look for a commit related to
>>>> tcp_tsq_limit_output_interval.  Try setting it to 5 or higher.
>>>
>>> This solved the problem for me.
>>>
>>> Link to commit for anyone else looking:
>>>
>>> https://github.com/greearb/linux-ct-4.13/commit/dc201c3c310f805681dbf393484bdc9715745df1
>>
>> Out of curiosity, did you try the patch that Kalle suggested?
>>
>> I'd be curious how that performed in comparison...
>>
>>
>> Thanks,
>> Ben
>>
>>
>> --
>> Ben Greear <greearb@candelatech.com>
>> Candela Technologies Inc  http://www.candelatech.com
>>
>
>

-- 
Mit freundlichen Grüssen / Regards

Sebastian Gottschall / CTO

NewMedia-NET GmbH - DD-WRT
Firmensitz:  Stubenwaldallee 21a, 64625 Bensheim
Registergericht: Amtsgericht Darmstadt, HRB 25473
Geschäftsführer: Peter Steinhäuser, Christian Scheele
http://www.dd-wrt.com
email: s.gottschall@dd-wrt.com
Tel.: +496251-582650 / Fax: +496251-5826565


_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k

Re: Bad TCP performance on WLE900VX

[Robert Jones]

>>> Out of curiosity, did you try the patch that Kalle suggested?
>>
>> I did not unfortunately. I have other cards to qualify but if I can
>> come back to this I'll post the results.
>
> this patch wont change anything, since you still need to modify ath10k to
> set a sane tsq value
> see this additional patch from lede
> https://git.lede-project.org/?p=openwrt/openwrt.git;a=commitdiff;h=111b49902465116a8353d29afe02eff0f56ea0a3

I was just planning on hardcoding a few values suggested by the
testing done in the Kalle's commit message.
But thank you for the LEDE patch, that will make testing this easier.

Thanks,
Robert

_______________________________________________
ath10k mailing list
ath10k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath10k