2012-05-29 22:57:14

by Ben Greear

[permalink] [raw]
Subject: Ath9k performance testing results (AR9380)

Ok, so thanks for all the suggestions and numbers that
folks have posted in the 'wifi throughput tests' thread.

Short answer: 352Mbps download, 270Mbps upload (not concurrent).


We set up two systems, both with WPEA-127N NICs (AR9380).

The Station machine is dual-core Atom, 3.3.7+ 32-bit kernel, with
Felix's recent optimizations and a bunch of other patches.

The AP is a core-2 DUO system running the same software. The
AP is set up to route.

Open air connection. Channel 149, HT-40. I can post hostapd
and supplicant config files if anyone wants to see them. AP
and STA machine are about 5 feet apart, turned so that antenna
face each other. With antennas on positioned so that they are
away from each other, performance was much worse.

The STAtion is sending to/from a wired port to/from the
station interface, so it is sending to itself.

Using ~64k UDP frames, these systems can sustain about 352Mbps
of traffic received on station interface and sent from the wired port (through
the AP). Our traffic generator cannot push 350Mbps to self when using
small-sized UDP frames on this hardware.

The traffic generator is our proprietary tool, since iperf can't easily
send to self, but I see no reason why iperf would be any slower if set up
properly with a third machine to act as the upstream iperf server.

When sending from STA to Wired, it maxes out at about 270Mbps.

I am not sure why there is such a big difference, but possibly
sending wifi is harder than receiving it, and the Atom processor
just can't keep up.

TCP is a lot slower than UDP..around 235Mbps download. I haven't
tried tuning things yet..maybe window sizes need some stretching.

We'll do some more tests with our i7 machines when we get
a chance...

Thanks!
Ben

--
Ben Greear <[email protected]>
Candela Technologies Inc http://www.candelatech.com



2012-05-31 02:49:35

by Sujith Manoharan

[permalink] [raw]
Subject: Ath9k performance testing results (AR9380)

Ben Greear wrote:
> I am not sure why there is such a big difference, but possibly
> sending wifi is harder than receiving it, and the Atom processor
> just can't keep up.

I have a small Atom-based netbook, I'll try with it and see what numbers
can be seen. It's a pain to pry it open and bolt a different card, though. :)

Sujith

2012-05-31 06:37:08

by Sujith Manoharan

[permalink] [raw]
Subject: Re: Ath9k performance testing results (AR9380)

Ben Greear wrote:
> On 05/30/2012 07:48 PM, Sujith Manoharan wrote:
> > Ben Greear wrote:
> >> I am not sure why there is such a big difference, but possibly
> >> sending wifi is harder than receiving it, and the Atom processor
> >> just can't keep up.
> >
> > I have a small Atom-based netbook, I'll try with it and see what numbers
> > can be seen. It's a pain to pry it open and bolt a different card, though. :)
>
> We had to install a hacked 'white-listed' BIOS to get a Lenovo laptop to even POST with
> an Atheros NIC in it, so if your NIC works w/out having to hack the BIOS, please
> let me know :)

It works on Samsung netbooks. I replaced the onboard card with a AR9462 card
and there were no hiccups.

Sujith

2012-05-31 05:18:26

by Ben Greear

[permalink] [raw]
Subject: Re: Ath9k performance testing results (AR9380)

On 05/30/2012 07:48 PM, Sujith Manoharan wrote:
> Ben Greear wrote:
>> I am not sure why there is such a big difference, but possibly
>> sending wifi is harder than receiving it, and the Atom processor
>> just can't keep up.
>
> I have a small Atom-based netbook, I'll try with it and see what numbers
> can be seen. It's a pain to pry it open and bolt a different card, though. :)

We had to install a hacked 'white-listed' BIOS to get a Lenovo laptop to even POST with
an Atheros NIC in it, so if your NIC works w/out having to hack the BIOS, please
let me know :)

Thanks,
Ben

--
Ben Greear <[email protected]>
Candela Technologies Inc http://www.candelatech.com