Question about ath10k throughput

Question about ath10k throughput

[Shu, Nick]

Hi, Group:
I'm doing throughout measurement on QCA9880 chip, running ath10k 3.17 driver with firmware.
I'm able to set channel width to 80MHz, enabled all three streams, MCS index to 9 for max. data throughput.
I verified all initial setting are correct.
But when test starts, MCS dropped (as the results of packet error?).
Is there any app command to send to driver to lock the MCS index during the test, so we can see better throughput?
What is the best practical number we can get for ath10k chip? 800? 900?

Thanks!
Nick Shu
Spirent Communications e-mail confidentiality.
------------------------------------------------------------------------
This e-mail contains confidential and / or privileged information belonging to Spirent Communications plc, its affiliates and / or subsidiaries. If you are not the intended recipient, you are hereby notified that any disclosure, copying, distribution and / or the taking of any action based upon reliance on the contents of this transmission is strictly forbidden. If you have received this message in error please notify the sender by return e-mail and delete it from your system.

Spirent Communications plc
Northwood Park, Gatwick Road, Crawley, West Sussex, RH10 9XN, United Kingdom.
Tel No. +44 (0) 1293 767676
Fax No. +44 (0) 1293 767677

Registered in England Number 470893
Registered at Northwood Park, Gatwick Road, Crawley, West Sussex, RH10 9XN, United Kingdom.

Or if within the US,

Spirent Communications,
27349 Agoura Road, Calabasas, CA, 91301, USA.
Tel No. 1-818-676- 2300

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

Re: Question about ath10k throughput

[Ben Greear]

On 04/21/2015 02:28 PM, Shu, Nick wrote:
> Hi, Group:
> I'm doing throughout measurement on QCA9880 chip, running ath10k 3.17 driver with firmware.
> I'm able to set channel width to 80MHz, enabled all three streams, MCS index to 9 for max. data throughput.
> I verified all initial setting are correct.
> But when test starts, MCS dropped (as the results of packet error?).
> Is there any app command to send to driver to lock the MCS index during the test, so we can see better throughput?
> What is the best practical number we can get for ath10k chip? 800? 900?

Please let us know more about your test setup:

Traffic protocol type (TCP, UDP, etc)

Over-the-air tests, or using cables + attenuators?

Is this in normal environment, or in an RF isolation chamber?

We've seen as much as 950Mbps UDP download using a certain Netgear AP,
but it is not reproducible.  We normally do this type of test using
sma cables + attenuators, and get mostly similar results whether in
isolation chamber or not when cabled up as long as the RF spectrum is
relatively quiet.

More often we see TCP in the 400-600 Mbps range and UDP in 700-800 Mbps
range..and with TCP, it can also be tricky to get good throughput
in our experience.

Increasing number of virtual stations decreases total throughput on most APs.

I have tried forcing the rate-control to run at specific (high) rates,
but it has not increased throughput in tests so far, so I have not played
with it that much.

Thanks,
Ben


> 
> Thanks!
> Nick Shu
> Spirent Communications e-mail confidentiality.
> ------------------------------------------------------------------------
> This e-mail contains confidential and / or privileged information belonging to Spirent Communications plc, its affiliates and / or subsidiaries. If you are not the intended recipient, you are hereby notified that any disclosure, copying, distribution and / or the taking of any action based upon reliance on the contents of this transmission is strictly forbidden. If you have received this message in error please notify the sender by return e-mail and delete it from your system.
> 
> Spirent Communications plc
> Northwood Park, Gatwick Road, Crawley, West Sussex, RH10 9XN, United Kingdom.
> Tel No. +44 (0) 1293 767676
> Fax No. +44 (0) 1293 767677
> 
> Registered in England Number 470893
> Registered at Northwood Park, Gatwick Road, Crawley, West Sussex, RH10 9XN, United Kingdom.
> 
> Or if within the US,
> 
> Spirent Communications,
> 27349 Agoura Road, Calabasas, CA, 91301, USA.
> Tel No. 1-818-676- 2300
> 
> _______________________________________________
> 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: Question about ath10k throughput

[Shu, Nick]

Thanks, Ben:
We use our test application to generate traffic (TCP or UDP), sending from wifi NIC (ONLY 1 CLIENT, not multiple vStas). RF cables are used to connect from NIC antenna port (x3) to Cisco AP (model#3702E) antenna port (AP is placed inside an isolation chamber). Scan results for that AP shows the signal strength is -18 dBm  (strong enough).
AP is connected back to test server through Ethernet cable, then Network host is route back the data packet to test application.
So the data throughput is calculated as: number of bytes * 2 (round trip) * 8 = Mbps.

So far, we can only get like 400-500 Mbps for UDP.
Should we set AP to open? Without encryption?

Thanks!
Nick





-----Original Message-----
From: Ben Greear [mailto:greearb@candelatech.com] 
Sent: Tuesday, April 21, 2015 4:38 PM
To: Shu, Nick
Cc: ath10k@lists.infradead.org
Subject: Re: Question about ath10k throughput

On 04/21/2015 02:28 PM, Shu, Nick wrote:
> Hi, Group:
> I'm doing throughout measurement on QCA9880 chip, running ath10k 3.17 driver with firmware.
> I'm able to set channel width to 80MHz, enabled all three streams, MCS index to 9 for max. data throughput.
> I verified all initial setting are correct.
> But when test starts, MCS dropped (as the results of packet error?).
> Is there any app command to send to driver to lock the MCS index during the test, so we can see better throughput?
> What is the best practical number we can get for ath10k chip? 800? 900?

Please let us know more about your test setup:

Traffic protocol type (TCP, UDP, etc)

Over-the-air tests, or using cables + attenuators?

Is this in normal environment, or in an RF isolation chamber?

We've seen as much as 950Mbps UDP download using a certain Netgear AP, but it is not reproducible.  We normally do this type of test using sma cables + attenuators, and get mostly similar results whether in isolation chamber or not when cabled up as long as the RF spectrum is relatively quiet.

More often we see TCP in the 400-600 Mbps range and UDP in 700-800 Mbps range..and with TCP, it can also be tricky to get good throughput in our experience.

Increasing number of virtual stations decreases total throughput on most APs.

I have tried forcing the rate-control to run at specific (high) rates, but it has not increased throughput in tests so far, so I have not played with it that much.

Thanks,
Ben


> 
> Thanks!
> Nick Shu
> Spirent Communications e-mail confidentiality.
> ----------------------------------------------------------------------
> -- This e-mail contains confidential and / or privileged information 
> belonging to Spirent Communications plc, its affiliates and / or subsidiaries. If you are not the intended recipient, you are hereby notified that any disclosure, copying, distribution and / or the taking of any action based upon reliance on the contents of this transmission is strictly forbidden. If you have received this message in error please notify the sender by return e-mail and delete it from your system.
> 
> Spirent Communications plc
> Northwood Park, Gatwick Road, Crawley, West Sussex, RH10 9XN, United Kingdom.
> Tel No. +44 (0) 1293 767676
> Fax No. +44 (0) 1293 767677
> 
> Registered in England Number 470893
> Registered at Northwood Park, Gatwick Road, Crawley, West Sussex, RH10 9XN, United Kingdom.
> 
> Or if within the US,
> 
> Spirent Communications,
> 27349 Agoura Road, Calabasas, CA, 91301, USA.
> Tel No. 1-818-676- 2300
> 
> _______________________________________________
> 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: Question about ath10k throughput

[Ben Greear]

On 04/21/2015 02:59 PM, Shu, Nick wrote:
> Thanks, Ben:
> We use our test application to generate traffic (TCP or UDP), sending from wifi NIC (ONLY 1 CLIENT, not multiple vStas). RF cables are used to connect from NIC antenna port (x3) to Cisco AP (model#3702E) antenna port (AP is placed inside an isolation chamber). Scan results for that AP shows the signal strength is -18 dBm  (strong enough).
> AP is connected back to test server through Ethernet cable, then Network host is route back the data packet to test application.
> So the data throughput is calculated as: number of bytes * 2 (round trip) * 8 = Mbps.
> 
> So far, we can only get like 400-500 Mbps for UDP.
> Should we set AP to open? Without encryption?

I don't see why you do the 2x to get 'round-trip'...I think for issues related
to wifi throughput, just talk about the amount of data flowing across the
wifi network interface itself.

Receiving on a station (download), when using CT firmware and with sw-crypt enabled,
does decryption on the CPU, and this is normally a CPU bound issue at around
500Mbps even on expensive E5 processors.

Upload uses HW encryption, so should run faster with modest host CPU usage.

I would try Open auth testing when verifying wifi throughput..then compare that
against your encrypted throughput to see if you are CPU bound (or, at least
not RF bound).

If you are doing true TCP protocol, then it will back off on drops (and require
TCP ack frames and so forth).  You will get quite a bit more bulk throughput
with UDP.

I would also try multiple different APs to try to find the best one.

Big expensive names on the AP don't always mean so much for basic throughput
tests in our experience.

Having the AP inside an isolation chamber is helpful, but you really
need your station system inside a chamber as well for a truly isolated test, otherwise
the station system may be picking up interference that causes lost frames
and so forth (and AP might pick up similar interference through cables
connected to your test equipment).

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: Question about ath10k throughput

[Shu, Nick]

Hi, Ben:
The data packet do go through the wifi interface twice, so I times 2.
Its hard to put DUT into the chamber, cause its too big.
I'm using your patched driver and firmware, that disabled HW encryption, so you said this will add CPU load?
Our application can be set to run traffic in one direction, so I will test again to check the throughout in downlink, uplink, and bidirection for comparison.
Will also try different AP, NetGear 1900 and Asus one.

Thanks!
Nick
________________________________________
From: Ben Greear [greearb@candelatech.com]
Sent: Tuesday, April 21, 2015 4:17 PM
To: Shu, Nick
Cc: ath10k@lists.infradead.org
Subject: Re: Question about ath10k throughput

On 04/21/2015 02:59 PM, Shu, Nick wrote:
> Thanks, Ben:
> We use our test application to generate traffic (TCP or UDP), sending from wifi NIC (ONLY 1 CLIENT, not multiple vStas). RF cables are used to connect from NIC antenna port (x3) to Cisco AP (model#3702E) antenna port (AP is placed inside an isolation chamber). Scan results for that AP shows the signal strength is -18 dBm  (strong enough).
> AP is connected back to test server through Ethernet cable, then Network host is route back the data packet to test application.
> So the data throughput is calculated as: number of bytes * 2 (round trip) * 8 = Mbps.
>
> So far, we can only get like 400-500 Mbps for UDP.
> Should we set AP to open? Without encryption?

I don't see why you do the 2x to get 'round-trip'...I think for issues related
to wifi throughput, just talk about the amount of data flowing across the
wifi network interface itself.

Receiving on a station (download), when using CT firmware and with sw-crypt enabled,
does decryption on the CPU, and this is normally a CPU bound issue at around
500Mbps even on expensive E5 processors.

Upload uses HW encryption, so should run faster with modest host CPU usage.

I would try Open auth testing when verifying wifi throughput..then compare that
against your encrypted throughput to see if you are CPU bound (or, at least
not RF bound).

If you are doing true TCP protocol, then it will back off on drops (and require
TCP ack frames and so forth).  You will get quite a bit more bulk throughput
with UDP.

I would also try multiple different APs to try to find the best one.

Big expensive names on the AP don't always mean so much for basic throughput
tests in our experience.

Having the AP inside an isolation chamber is helpful, but you really
need your station system inside a chamber as well for a truly isolated test, otherwise
the station system may be picking up interference that causes lost frames
and so forth (and AP might pick up similar interference through cables
connected to your test equipment).

Thanks,
Ben


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

Spirent Communications e-mail confidentiality.
------------------------------------------------------------------------
This e-mail contains confidential and / or privileged information belonging to Spirent Communications plc, its affiliates and / or subsidiaries. If you are not the intended recipient, you are hereby notified that any disclosure, copying, distribution and / or the taking of any action based upon reliance on the contents of this transmission is strictly forbidden. If you have received this message in error please notify the sender by return e-mail and delete it from your system.

Spirent Communications plc
Northwood Park, Gatwick Road, Crawley, West Sussex, RH10 9XN, United Kingdom.
Tel No. +44 (0) 1293 767676
Fax No. +44 (0) 1293 767677

Registered in England Number 470893
Registered at Northwood Park, Gatwick Road, Crawley, West Sussex, RH10 9XN, United Kingdom.

Or if within the US,

Spirent Communications,
27349 Agoura Road, Calabasas, CA, 91301, USA.
Tel No. 1-818-676- 2300

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

Re: Question about ath10k throughput

[Ben Greear]

On 04/21/2015 04:07 PM, Shu, Nick wrote:
> Hi, Ben:
> The data packet do go through the wifi interface twice, so I times 2.
> Its hard to put DUT into the chamber, cause its too big.
> I'm using your patched driver and firmware, that disabled HW encryption, so you said this will add CPU load?
> Our application can be set to run traffic in one direction, so I will test again to check the throughout in downlink, uplink, and bidirection for comparison.
> Will also try different AP, NetGear 1900 and Asus one.

In general, bi-directional traffic on wifi is a lot slower than uni-directional,
even if you add up both directions in bi-directional traffic.

I think this is mostly because the AP and station radios contend for airtime, and
you end up taking less advantage of block-acks and A-MPDUS.  Maybe more retransmits
and backoffs too..  Maybe others will have a better idea.

Disabling HW encryption makes the CPU do the decryption, so yeah, that is CPU intensive.

'perf top' or similar should show your cpu usage hotspots.

Thanks,
Ben



> 
> Thanks!
> Nick
> ________________________________________
> From: Ben Greear [greearb@candelatech.com]
> Sent: Tuesday, April 21, 2015 4:17 PM
> To: Shu, Nick
> Cc: ath10k@lists.infradead.org
> Subject: Re: Question about ath10k throughput
> 
> On 04/21/2015 02:59 PM, Shu, Nick wrote:
>> Thanks, Ben:
>> We use our test application to generate traffic (TCP or UDP), sending from wifi NIC (ONLY 1 CLIENT, not multiple vStas). RF cables are used to connect from NIC antenna port (x3) to Cisco AP (model#3702E) antenna port (AP is placed inside an isolation chamber). Scan results for that AP shows the signal strength is -18 dBm  (strong enough).
>> AP is connected back to test server through Ethernet cable, then Network host is route back the data packet to test application.
>> So the data throughput is calculated as: number of bytes * 2 (round trip) * 8 = Mbps.
>>
>> So far, we can only get like 400-500 Mbps for UDP.
>> Should we set AP to open? Without encryption?
> 
> I don't see why you do the 2x to get 'round-trip'...I think for issues related
> to wifi throughput, just talk about the amount of data flowing across the
> wifi network interface itself.
> 
> Receiving on a station (download), when using CT firmware and with sw-crypt enabled,
> does decryption on the CPU, and this is normally a CPU bound issue at around
> 500Mbps even on expensive E5 processors.
> 
> Upload uses HW encryption, so should run faster with modest host CPU usage.
> 
> I would try Open auth testing when verifying wifi throughput..then compare that
> against your encrypted throughput to see if you are CPU bound (or, at least
> not RF bound).
> 
> If you are doing true TCP protocol, then it will back off on drops (and require
> TCP ack frames and so forth).  You will get quite a bit more bulk throughput
> with UDP.
> 
> I would also try multiple different APs to try to find the best one.
> 
> Big expensive names on the AP don't always mean so much for basic throughput
> tests in our experience.
> 
> Having the AP inside an isolation chamber is helpful, but you really
> need your station system inside a chamber as well for a truly isolated test, otherwise
> the station system may be picking up interference that causes lost frames
> and so forth (and AP might pick up similar interference through cables
> connected to your test equipment).
> 
> Thanks,
> Ben
> 
> 
> --
> Ben Greear <greearb@candelatech.com>
> Candela Technologies Inc  http://www.candelatech.com
> 
> Spirent Communications e-mail confidentiality.
> ------------------------------------------------------------------------
> This e-mail contains confidential and / or privileged information belonging to Spirent Communications plc, its affiliates and / or subsidiaries. If you are not the intended recipient, you are hereby notified that any disclosure, copying, distribution and / or the taking of any action based upon reliance on the contents of this transmission is strictly forbidden. If you have received this message in error please notify the sender by return e-mail and delete it from your system.
> 
> Spirent Communications plc
> Northwood Park, Gatwick Road, Crawley, West Sussex, RH10 9XN, United Kingdom.
> Tel No. +44 (0) 1293 767676
> Fax No. +44 (0) 1293 767677
> 
> Registered in England Number 470893
> Registered at Northwood Park, Gatwick Road, Crawley, West Sussex, RH10 9XN, United Kingdom.
> 
> Or if within the US,
> 
> Spirent Communications,
> 27349 Agoura Road, Calabasas, CA, 91301, USA.
> Tel No. 1-818-676- 2300
> 


-- 
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: Question about ath10k throughput

[Michal Kazior]

On 21 April 2015 at 23:28, Shu, Nick <Nick.Shu@spirent.com> wrote:
> Hi, Group:
> I'm doing throughout measurement on QCA9880 chip, running ath10k 3.17 driver with firmware.
> I'm able to set channel width to 80MHz, enabled all three streams, MCS index to 9 for max. data throughput.
> I verified all initial setting are correct.
> But when test starts, MCS dropped (as the results of packet error?).

> Is there any app command to send to driver to lock the MCS index during the test, so we can see better throughput?

You can try:

  iw wlan0 set bitrates legacy-5 ht-mcs-5 vht-mcs-5 3:9

To reset:

  iw wlan0 set bitrates


> What is the best practical number we can get for ath10k chip? 800? 900?

On a cabled RF with attenuators I remember seeing ~900mbps of UDP
between two QCA988Xs and ~650mbps of UDP between QCA988X and QCA6174,


Michał

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