Cannot achieve throughput higher than 460Mbps using CT firmware and 3x3 ath10k

Cannot achieve throughput higher than 460Mbps using CT firmware and 3x3 ath10k

[richard mayers]

[-- Attachment #1: Type: text/plain, Size: 1338 bytes --]

Hi all,

The maximum throughput I get with the following set up is around
460-470Mbits/s when using MCS 6, 80MHz channel and 3 spatial streams.
However, when I try MCS 7, the throughput drops to 200Mbits/s, and
when I use 8 or 9 it goes to 0. The signal level is -11dBm which I
guess is quite good.

Do you have any indications of what I do wrong and cannot get 600Mbps
throughput, as mention in previous messages in this list? Any help is
highly appreciated.

My set up is as follows:

- STA: compex wle900vx (QCA988x), ubuntu 14.04 LTS, dell dual core
1.66ghz, kernel version 3.16, driver backport 4.0.1 and as a firmware
the commercial firmware from Candela Tech (
10.4.467-ct-com-full-013-b5b14a).

- AP: Has the same set up, but is running hostapd with the
configuration file that you can find attached.

- I tried with many distances, but to make it easier now I have them
placed 50cm from each other (with no obstacles in between), with
3-antenna array and each antenna separated 12 cm from each other.

- For the throughput tests I use iperf (UDP). I run the server in the
AP, and the client in the station (upload). Command used in the client
: "iperf -c <IP> -i 1 -b 600m -S 0x08".

I also tried different drivers, kernels, firmware, distances between
STA and AP, kernel sockets buffers, etc.. But I can not get better
results.

[-- Attachment #2: hostapd80.conf --]
[-- Type: application/octet-stream, Size: 633 bytes --]

driver=nl80211
logger_syslog=127
logger_syslog_level=2
logger_stdout=127
logger_stdout_level=2
hw_mode=a
channel=149

ieee80211n=1
ht_capab=[HT40+][LDPC][SHORT-GI-20][SHORT-GI-40][TX-STBC][RX-STBC1][MAX-AMSDU-7935][DSSS_CCK-40]
vht_oper_chwidth=1
vht_oper_centr_freq_seg0_idx=155
ieee80211ac=1
vht_capab=[RXLDPC][SHORT-GI-80][TX-STBC-2BY1][RX-ANTENNA-PATTERN][TX-ANTENNA-PATTERN][RX-STBC1][MAX-MPDU-11454][MAX-A-MPDU-LEN-EXP7]

interface=wlan4
ctrl_interface=/var/run/hostapd
disassoc_low_ack=1
preamble=1
wmm_enabled=1
ignore_broadcast_ssid=0
uapsd_advertisement_enabled=1
auth_algs=1
wpa=0
ssid=wirelessAC
bssid=04:f0:21:11:e6:76


[-- Attachment #3: Type: text/plain, Size: 146 bytes --]

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

Re: Cannot achieve throughput higher than 460Mbps using CT firmware and 3x3 ath10k

[Ben Greear]

On 05/06/2015 08:19 AM, Tomasz Skapski wrote:
> Hi,
> 
> I am sure that it does not solve your problem with MCS 8 and 9, but to get better throughput using iperf, run few streams in parallel: -P X.
> 
> eg:
> iperf -c <IP> -i1 -u -b 200M -P4

Larger kernel buffer sizes (1MB, for instance), and larger PDU writes (9k or maybe 24k) might help as well.

There is some wifi tuning info here...the details are mostly relating to our
traffic generator, but the general idea applies to iperf and other
traffic generators as well...

http://www.candelatech.com/lf_wifi_perf.php


Thanks,
Ben

> 
> 
> 
> 
> Regards,
> Tomasz Skąpski
> 
> On 6 May 2015 at 17:01, richard mayers <richard.mayers92@gmail.com <mailto:richard.mayers92@gmail.com>> wrote:
> 
>     Hi all,
> 
>     The maximum throughput I get with the following set up is around
>     460-470Mbits/s when using MCS 6, 80MHz channel and 3 spatial streams.
>     However, when I try MCS 7, the throughput drops to 200Mbits/s, and
>     when I use 8 or 9 it goes to 0. The signal level is -11dBm which I
>     guess is quite good.
> 
>     Do you have any indications of what I do wrong and cannot get 600Mbps
>     throughput, as mention in previous messages in this list? Any help is
>     highly appreciated.
> 
>     My set up is as follows:
> 
>     - STA: compex wle900vx (QCA988x), ubuntu 14.04 LTS, dell dual core
>     1.66ghz, kernel version 3.16, driver backport 4.0.1 and as a firmware
>     the commercial firmware from Candela Tech (
>     10.4.467-ct-com-full-013-b5b14a).
> 
>     - AP: Has the same set up, but is running hostapd with the
>     configuration file that you can find attached.
> 
>     - I tried with many distances, but to make it easier now I have them
>     placed 50cm from each other (with no obstacles in between), with
>     3-antenna array and each antenna separated 12 cm from each other.
> 
>     - For the throughput tests I use iperf (UDP). I run the server in the
>     AP, and the client in the station (upload). Command used in the client
>     : "iperf -c <IP> -i 1 -b 600m -S 0x08".
> 
>     I also tried different drivers, kernels, firmware, distances between
>     STA and AP, kernel sockets buffers, etc.. But I can not get better
>     results.
> 
>     _______________________________________________
>     ath10k mailing list
>     ath10k@lists.infradead.org <mailto:ath10k@lists.infradead.org>
>     http://lists.infradead.org/mailman/listinfo/ath10k
> 
> 


-- 
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: Cannot achieve throughput higher than 460Mbps using CT firmware and 3x3 ath10k

[Michal Kazior]

On 6 May 2015 at 17:01, richard mayers <richard.mayers92@gmail.com> wrote:
> Hi all,
>
> The maximum throughput I get with the following set up is around
> 460-470Mbits/s when using MCS 6, 80MHz channel and 3 spatial streams.
> However, when I try MCS 7, the throughput drops to 200Mbits/s, and
> when I use 8 or 9 it goes to 0. The signal level is -11dBm which I
> guess is quite good.

Hmm.. From my experience too strong signal can degrade performance
with qca988x. You could try reducing it via `iw`. You could also try
checking if using only 2 spatial streams also yields problems with
MCS8 and MCS9.

Antenna themselves might be the problem as well. Strong signal doesn't
mean good signal quality. If your antenna connector is damaged or
antenna is of poor quality you might end up not being able to use
highest MCS rates.


Michał

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

Fwd: Cannot achieve throughput higher than 460Mbps using CT firmware and 3x3 ath10k

[richard mayers]

---------- Forwarded message ----------
From: richard mayers <richard.mayers92@gmail.com>
Date: 2015-05-07 9:51 GMT+02:00
Subject: Re: Cannot achieve throughput higher than 460Mbps using CT
firmware and 3x3 ath10k
To: Michal Kazior <michal.kazior@tieto.com>


Hi,

Yes, with only one spatial stream I can use MCS9 and I get 290Mbps
more or less, then with 2 spatial streams I can also use MCS 8, but as
I said with 3 streams  I only get a "good throughput" with  MC6.

Richard

2015-05-07 7:47 GMT+02:00 Michal Kazior <michal.kazior@tieto.com>:
> On 6 May 2015 at 17:01, richard mayers <richard.mayers92@gmail.com> wrote:
>> Hi all,
>>
>> The maximum throughput I get with the following set up is around
>> 460-470Mbits/s when using MCS 6, 80MHz channel and 3 spatial streams.
>> However, when I try MCS 7, the throughput drops to 200Mbits/s, and
>> when I use 8 or 9 it goes to 0. The signal level is -11dBm which I
>> guess is quite good.
>
> Hmm.. From my experience too strong signal can degrade performance
> with qca988x. You could try reducing it via `iw`. You could also try
> checking if using only 2 spatial streams also yields problems with
> MCS8 and MCS9.
>
> Antenna themselves might be the problem as well. Strong signal doesn't
> mean good signal quality. If your antenna connector is damaged or
> antenna is of poor quality you might end up not being able to use
> highest MCS rates.
>
>
> Michał

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

Re: Fwd: Cannot achieve throughput higher than 460Mbps using CT firmware and 3x3 ath10k

[Ben Greear]

You tried your antenna at different orientations?  Seems they work better
when they are somewhat orthogonal to each other.

Thanks,
Ben


On 05/07/2015 12:52 AM, richard mayers wrote:
> ---------- Forwarded message ----------
> From: richard mayers <richard.mayers92@gmail.com>
> Date: 2015-05-07 9:51 GMT+02:00
> Subject: Re: Cannot achieve throughput higher than 460Mbps using CT
> firmware and 3x3 ath10k
> To: Michal Kazior <michal.kazior@tieto.com>
>
>
> Hi,
>
> Yes, with only one spatial stream I can use MCS9 and I get 290Mbps
> more or less, then with 2 spatial streams I can also use MCS 8, but as
> I said with 3 streams  I only get a "good throughput" with  MC6.
>
> Richard
>
> 2015-05-07 7:47 GMT+02:00 Michal Kazior <michal.kazior@tieto.com>:
>> On 6 May 2015 at 17:01, richard mayers <richard.mayers92@gmail.com> wrote:
>>> Hi all,
>>>
>>> The maximum throughput I get with the following set up is around
>>> 460-470Mbits/s when using MCS 6, 80MHz channel and 3 spatial streams.
>>> However, when I try MCS 7, the throughput drops to 200Mbits/s, and
>>> when I use 8 or 9 it goes to 0. The signal level is -11dBm which I
>>> guess is quite good.
>>
>> Hmm.. From my experience too strong signal can degrade performance
>> with qca988x. You could try reducing it via `iw`. You could also try
>> checking if using only 2 spatial streams also yields problems with
>> MCS8 and MCS9.
>>
>> Antenna themselves might be the problem as well. Strong signal doesn't
>> mean good signal quality. If your antenna connector is damaged or
>> antenna is of poor quality you might end up not being able to use
>> highest MCS rates.
>>
>>
>> Michał
>
> _______________________________________________
> ath10k mailing list
> ath10k@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/ath10k
>

-- 
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: Fwd: Cannot achieve throughput higher than 460Mbps using CT firmware and 3x3 ath10k

[richard mayers]

If you mean orthogonal between receiver and transmitter I already
tried and same results.

I also tried to lower the transmission power from 30dbm to 18, 6, 12..
but nothing changed.

Thanks,
Richard

2015-05-07 16:19 GMT+02:00 Ben Greear <greearb@candelatech.com>:
> You tried your antenna at different orientations?  Seems they work better
> when they are somewhat orthogonal to each other.
>
> Thanks,
> Ben
>
>
>
> On 05/07/2015 12:52 AM, richard mayers wrote:
>>
>> ---------- Forwarded message ----------
>> From: richard mayers <richard.mayers92@gmail.com>
>> Date: 2015-05-07 9:51 GMT+02:00
>> Subject: Re: Cannot achieve throughput higher than 460Mbps using CT
>> firmware and 3x3 ath10k
>> To: Michal Kazior <michal.kazior@tieto.com>
>>
>>
>> Hi,
>>
>> Yes, with only one spatial stream I can use MCS9 and I get 290Mbps
>> more or less, then with 2 spatial streams I can also use MCS 8, but as
>> I said with 3 streams  I only get a "good throughput" with  MC6.
>>
>> Richard
>>
>> 2015-05-07 7:47 GMT+02:00 Michal Kazior <michal.kazior@tieto.com>:
>>>
>>> On 6 May 2015 at 17:01, richard mayers <richard.mayers92@gmail.com>
>>> wrote:
>>>>
>>>> Hi all,
>>>>
>>>> The maximum throughput I get with the following set up is around
>>>> 460-470Mbits/s when using MCS 6, 80MHz channel and 3 spatial streams.
>>>> However, when I try MCS 7, the throughput drops to 200Mbits/s, and
>>>> when I use 8 or 9 it goes to 0. The signal level is -11dBm which I
>>>> guess is quite good.
>>>
>>>
>>> Hmm.. From my experience too strong signal can degrade performance
>>> with qca988x. You could try reducing it via `iw`. You could also try
>>> checking if using only 2 spatial streams also yields problems with
>>> MCS8 and MCS9.
>>>
>>> Antenna themselves might be the problem as well. Strong signal doesn't
>>> mean good signal quality. If your antenna connector is damaged or
>>> antenna is of poor quality you might end up not being able to use
>>> highest MCS rates.
>>>
>>>
>>> Michał
>>
>>
>> _______________________________________________
>> ath10k mailing list
>> ath10k@lists.infradead.org
>> http://lists.infradead.org/mailman/listinfo/ath10k
>>
>
> --
> 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: Fwd: Cannot achieve throughput higher than 460Mbps using CT firmware and 3x3 ath10k

[Ben Greear]

On 05/07/2015 09:04 AM, richard mayers wrote:
> If you mean orthogonal between receiver and transmitter I already
> tried and same results.
> 
> I also tried to lower the transmission power from 30dbm to 18, 6, 12..
> but nothing changed.


Assuming you have 3 antenna on each system.  Put two vertical on both AP and
station.  Put third horizontal (say, pointing east) on each system.  Make sure
the units are turned such that the east antenna on the two sysetms
are parallel with each other.

You might then try changing one of the verticals to be a bit offset as well
(maybe 45 deg from vertical, matching orientation with peer device).

Finally, you need to move to the country and test on a farm with all household
appliances powered off and no other house within 1/2 mile.  Make sure there are no
airports or other radar towers within 50 miles.  Then, maybe you will have clean
environment :)

You can also buy a $20,000 isolation chamber, which will help a bit.

On the cheap end, old metal filing cabinets provide a bit of isolation, and
inside a refrigerator might help as well.  I would be curious if being inside a turned-off
microwave also helped, in case someone has done such tests!

All that said, even in isolation chambers we see funky throughput on ath10k,
especially in 3.17 and newer kernels (our old hacked-upon 3.14 kernel seems
to work the most reliable for throughput, but I don't yet know why that is the case,
I suspect buffer-bloat related patches in newer kernels is causing issues on bulk
throughput tests, though mostly we see these problems with tcp traffic, not UDP).

Thanks,
Ben


> 
> Thanks,
> Richard
> 
> 2015-05-07 16:19 GMT+02:00 Ben Greear <greearb@candelatech.com>:
>> You tried your antenna at different orientations?  Seems they work better
>> when they are somewhat orthogonal to each other.
>>
>> Thanks,
>> Ben
>>
>>
>>
>> On 05/07/2015 12:52 AM, richard mayers wrote:
>>>
>>> ---------- Forwarded message ----------
>>> From: richard mayers <richard.mayers92@gmail.com>
>>> Date: 2015-05-07 9:51 GMT+02:00
>>> Subject: Re: Cannot achieve throughput higher than 460Mbps using CT
>>> firmware and 3x3 ath10k
>>> To: Michal Kazior <michal.kazior@tieto.com>
>>>
>>>
>>> Hi,
>>>
>>> Yes, with only one spatial stream I can use MCS9 and I get 290Mbps
>>> more or less, then with 2 spatial streams I can also use MCS 8, but as
>>> I said with 3 streams  I only get a "good throughput" with  MC6.
>>>
>>> Richard
>>>
>>> 2015-05-07 7:47 GMT+02:00 Michal Kazior <michal.kazior@tieto.com>:
>>>>
>>>> On 6 May 2015 at 17:01, richard mayers <richard.mayers92@gmail.com>
>>>> wrote:
>>>>>
>>>>> Hi all,
>>>>>
>>>>> The maximum throughput I get with the following set up is around
>>>>> 460-470Mbits/s when using MCS 6, 80MHz channel and 3 spatial streams.
>>>>> However, when I try MCS 7, the throughput drops to 200Mbits/s, and
>>>>> when I use 8 or 9 it goes to 0. The signal level is -11dBm which I
>>>>> guess is quite good.
>>>>
>>>>
>>>> Hmm.. From my experience too strong signal can degrade performance
>>>> with qca988x. You could try reducing it via `iw`. You could also try
>>>> checking if using only 2 spatial streams also yields problems with
>>>> MCS8 and MCS9.
>>>>
>>>> Antenna themselves might be the problem as well. Strong signal doesn't
>>>> mean good signal quality. If your antenna connector is damaged or
>>>> antenna is of poor quality you might end up not being able to use
>>>> highest MCS rates.
>>>>
>>>>
>>>> Michał
>>>
>>>
>>> _______________________________________________
>>> ath10k mailing list
>>> ath10k@lists.infradead.org
>>> http://lists.infradead.org/mailman/listinfo/ath10k
>>>
>>
>> --
>> Ben Greear <greearb@candelatech.com>
>> Candela Technologies Inc  http://www.candelatech.com
> 


-- 
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: Cannot achieve throughput higher than 460Mbps using CT firmware and 3x3 ath10k

[Pushpal Sidhu]

On Wed, May 6, 2015 at 8:01 AM, richard mayers
<richard.mayers92@gmail.com> wrote:
> Hi all,
>
> The maximum throughput I get with the following set up is around
> 460-470Mbits/s when using MCS 6, 80MHz channel and 3 spatial streams.
> However, when I try MCS 7, the throughput drops to 200Mbits/s, and
> when I use 8 or 9 it goes to 0. The signal level is -11dBm which I
> guess is quite good.
>
> Do you have any indications of what I do wrong and cannot get 600Mbps
> throughput, as mention in previous messages in this list? Any help is
> highly appreciated.
>
> My set up is as follows:
>
> - STA: compex wle900vx (QCA988x), ubuntu 14.04 LTS, dell dual core
> 1.66ghz, kernel version 3.16, driver backport 4.0.1 and as a firmware
> the commercial firmware from Candela Tech (
> 10.4.467-ct-com-full-013-b5b14a).

Honestly, I think it's your CPU. Run top while doing your throughput
test, and I think you'll see 100% sirq (pegged CPU core). Also make
sure you split the CPU's up so you're not looking at the 'total' cpu
count.

- Pushpal

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

Re: Cannot achieve throughput higher than 460Mbps using CT firmware and 3x3 ath10k

[richard mayers]

Its only 30% of CPU usage when I am transmit at maximum speed. May be
I am not checking it correctly, I just use top and then press 1 and it
splits the cores (it has only 2). In the worst case if 30% was the
average it should be a maximum of 60% in one single core.

Richard.

2015-05-07 19:00 GMT+02:00 Pushpal Sidhu <psidhu@gateworks.com>:
> On Wed, May 6, 2015 at 8:01 AM, richard mayers
> <richard.mayers92@gmail.com> wrote:
>> Hi all,
>>
>> The maximum throughput I get with the following set up is around
>> 460-470Mbits/s when using MCS 6, 80MHz channel and 3 spatial streams.
>> However, when I try MCS 7, the throughput drops to 200Mbits/s, and
>> when I use 8 or 9 it goes to 0. The signal level is -11dBm which I
>> guess is quite good.
>>
>> Do you have any indications of what I do wrong and cannot get 600Mbps
>> throughput, as mention in previous messages in this list? Any help is
>> highly appreciated.
>>
>> My set up is as follows:
>>
>> - STA: compex wle900vx (QCA988x), ubuntu 14.04 LTS, dell dual core
>> 1.66ghz, kernel version 3.16, driver backport 4.0.1 and as a firmware
>> the commercial firmware from Candela Tech (
>> 10.4.467-ct-com-full-013-b5b14a).
>
> Honestly, I think it's your CPU. Run top while doing your throughput
> test, and I think you'll see 100% sirq (pegged CPU core). Also make
> sure you split the CPU's up so you're not looking at the 'total' cpu
> count.
>
> - Pushpal

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

Re: Cannot achieve throughput higher than 460Mbps using CT firmware and 3x3 ath10k

[Pushpal Sidhu]

On Thu, May 7, 2015 at 11:01 AM, richard mayers
<richard.mayers92@gmail.com> wrote:
> Its only 30% of CPU usage when I am transmit at maximum speed. May be
> I am not checking it correctly, I just use top and then press 1 and it
> splits the cores (it has only 2). In the worst case if 30% was the
> average it should be a maximum of 60% in one single core.

And you're getting your 30% CPU usage from (100 - IDLE), correct?
Don't just look at the 'us' column.

Going back to over drive on the chains, instead of mucking around with
antenna orientation and distance, I would go out and buy some
attenuators and put them between the channels. Then change the tx
power on the AP until you're absolutely tuned up. Otherwise, these
tests are somewhat moot in my opinion.

- Pushpal

> Richard.

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