Tuesday, April 9, 2013

First Look - 802.11ac Gigabit Wi-Fi Speed

I just couldn't wait for 802.11ac enterprise WLAN gear to be released, so I went and picked up a consumer router and USB adapter to get a first look at 802.11ac.

Equipment
I picked up the ASUS RT-AC66U and corresponding ASUS USB-AC53 adapter. I decided to go this route because with pre-standard equipment we've historically seen best performance when pairing the AP and client adapters from the same manufacturer.

ASUS 802.11ac Product Lineup

The router supports 3x3:3 MIMO with 80 MHz channels and 256-QAM on the 5GHz 802.11ac capable radio. This results in a raw Wi-Fi data rate capable of 1.3 Gbps. You might want to reference this 802.11ac data rate chart.

The USB adapter on the other hand is a bit handicapped, supporting only 2x2:2 MIMO, 80 MHz channels, and 256-QAM. This results in a raw Wi-Fi data rate capable of 867 Mbps. However, actual throughput performance is further limited by it's USB 2.0 interface which has a raw bus speed of 480Mbps but is limited to somewhere around 60% of that due to USB host controller overhead. So realistically, with this USB adapter I'm going to max out around 288 Mbps of actual throughput. Additionally, it's increasingly hard to pack a 3x3:3 MIMO antenna system into an external USB adapter due to physical size constraints. This is what we've seen with external 802.11n adapters for the most part, and that has continued to be the case with most first-generation external 802.11ac client adapters. We should see more spatial stream support on internal Wi-Fi adapters where the antenna can be integrated into the laptop case in order to create sufficient physical separation.

I loaded the adapter driver and utility into a Windows 7 laptop and it reported the raw speed as 866.5 Mbps, which is equivalent to the adapter's maximum raw Wi-Fi speed based on its specifications.

Windows Connection Details Reported an 866.5 Mbps Speed

Peak Performance
I decided to initially perform a simple file transfer using SMB. To accomplish this, I hooked one laptop to the wired Gigabit Ethernet LAN port on the router as the server and connected the second laptop to the 5 GHz Wi-Fi radio as the client. I pulled a 1.38GB file down from the server to the client over the wireless link. I performed this test over 10 times for reproducibility. The router was configured for 5 GHZ operation with an 80 MHz channel and a primary channel of 48.

File size: 1.38GB / 1,490,209,655 Bytes / 11,921,677,240 bits
Transfer time: 51.0 seconds (avg.)
Application throughput: 233.76 Mbps

Note - remember that this is application layer throughput, which does not include SMB, TCP, IP, or MAC layer overhead.

Next, I performed an iPerf TCP throughput transfer for 60 seconds, modifying the TCP window size to 1024KB. The test was performed ten times, with the average result clocking in at 204 Mbps.

iperf -f m -i 1 -w 1024K -c 192.168.1.33 -t 60

[ ID] Interval       Transfer     Bandwidth
[156]  0.0-60.0 sec  1482 MBytes   204 Mbits/sec

Overall, I was fairly pleased with this result. Although it is nowhere near the Gigabit speeds that 802.11ac is capable of achieving, this is to be expected given the handicapped USB 2.0 adapter that I am using. Given a raw data rate of 867 Mbps peak, I would expect to see throughput around 520 Mbps without the USB 2.0 limitation (or roughly 60% of the peak data rate).

I used MetaGeek's Chanalyzer Pro and WiSpy DBx to record the 80 MHz file transfers. I configured the router to use an 80 MHz channel width, with a primary channel of 48 in the UNII-1 band. The spectrum analysis workstation was located within 5 feet of the router.

Spectrum Analysis of an 802.11ac 80 MHz Channel
You can clearly see FFT waveform across the entire UNII-1 band due to the automatic extension of the primary 20 MHz channel up to the maximum 80 MHz channel width that was configured. As a recap of how the primary and extension channels work, you might want to review my previous post on 802.11ac Channel Planning.

Performance Comparison versus 802.11n
To assess how well the 2 spatial stream 802.11ac client performed, I decided to benchmark it against similarly capable 2SS 802.11n clients. To make the test apples-to-apples, I also turned down the channel width on the ASUS router to 40 MHz, since that is the largest channel width the 802.11n clients support.

I ran the same SMB file transfer test as described previously. The SMB file transfer was run three times for each client and I averaged the results. Each client was placed in the same physical location, approximately 10 feet from the router to assess peak performance.

40 MHz throughput comparison against two different 2x2:2 802.11n clients:

Wi-Fi AdapterWi-Fi CapabilitiesMax Wi-Fi Data Rate at 40 MHzSMB File Transfer TimeSMB Throughput
MacBook Air Airport (Internal)802.11n, 2x2:2 MIMO300 Mbps1:36.7 sec123.29 Mbps
Intel 4965ABGN (Internal)802.11n, 2x2:2 MIMO300 Mbps1:15.5 sec157.90 Mbps
Asus USB-AC53802.11ac, 2x2:2 MIMO400 Mbps
(limited further due to USB 2.0 bus speed)
1:07.0 sec177.94 Mbps

Clearly, the peak performance test shows that the 802.11ac client has an edge due to the higher data rates provided at the top-end with 256-QAM modulation.

Rate over Range versus 802.11n
The aggressive 256-QAM modulation may provide higher peak throughput when a client is physically very close to the AP. But will that advantage hold up over larger distances? There has been significant discussion in the industry about the real-world usefulness of 256-QAM, especially at typical client distances of 10-20 feet or greater from an AP. Will clients realistically be able to use such aggressive modulation in practice?

However, the question about the use of higher modulation is not the only one. Newer wireless chipsets should also benefit from better manufacturing processes that improve 802.11ac client receive sensitivity, translating into the use of higher data rates at at any given distance when compared to older 802.11n clients. Put another way, does 802.11ac exhibit better rate-over-range compared to 802.11n?

For the rate-over-range testing, I ran the same iPerf TCP throughput test as described previously, using a 40 MHz channel width at varying distances from the router. The tests were performed in a residential house since that is the only space that I have available at the moment.

- Point #1: 10 feet from the router, 1 wood panel wall in-between
- Point #2: 25 feet from the router, 1 wood panel wall in-between
- Point #3: 30 feet from the router, 1 wood floor and 1 drywall in-between
- Point #4: 40 feet from the router, 2 wood floors and 1 wood panel wall in-between

Additionally, in parenthesis I provide a rough signal strength and data rate used during the tests as reported by the client supplicant/driver. I also provided the test results for the 802.11ac client using an 80 MHz channel width for reference.

LocationMacbook Air 
(40 MHz)
Intel 4965ABGN 
(40 MHz)
ASUS USB-AC53 
(40 MHz)
ASUS USB-AC53 
(80 MHz)
Point 1154 Mbps
(-50 dBm, 300 Mbps)
165 Mbps
(-50 dBm, 300 Mbps)
204 Mbps 
(-40 dBm, 400 Mbps)
204 Mbps 
(-40 dBm, 867 Mbps)
Point 2144 Mbps
(-60 dBm, 216 Mbps)
150 Mbps
(-60 dBm, 270 Mbps)
180 Mbps 
(-52 dBm, 400 and 324 Mbps)
202 Mbps 
(-54 dBm, 702 and 585 Mbps)
Point 3112 Mbps
(-65 dBm, 162 Mbps)
102 Mbps
(-70 dBm, 180 Mbps)
144 Mbps 
(-54 dBm, 243 Mbps)
201 Mbps 
(-60 dBm, 526 Mbps)
Point 435 Mbps
(-80 dBm, 54 Mbps)
61 Mbps
(-80 dBm, 120 Mbps)
119 Mbps 
(-65 dBm, 216 Mbps)
190 Mbps 
(-70 dBm, 468 Mbps)

The 802.11ac client performance beats both 802.11n clients in all tests at all locations. Additionally, the performance gap widens significantly at Point #4, the farthest distance and weakest signal from the router. Clearly, 802.11ac provides significant rate-over-range improvements over 802.11n.

Performance improvement at each location:
- Point #1: 32.5% (vs MBA), 23.6% (vs Intel)
- Point #2: 25.0% (vs MBA), 20.0% (vs Intel)
- Point #3: 28.6% (vs MBA), 41.2% (vs Intel)
- Point #4: 240% (vs MBA), 95.1% (vs Intel)

The peak performance improvement at Point #1 (32.5%, 23.6%) is squarely in-line with 802.11ac's 33% theoretical improvement over 802.11n due to the higher modulation rate of 256-QAM (400 Mbps) versus 64-QAM (300 Mbps).

The receive sensitivity of the 802.11ac client is also better than the 802.11n clients. At most test locations the 11ac adapter exhibited an 8-15 dB signal advantage over the older 11n adapters. This highlights the fact that newer wireless chipsets offer improved hardware quality over older chipsets. Receive sensitivity also appears to be better for the same 802.11ac adapter when smaller channel widths are used. This highlights the fact that as channel width increases, clients will need to maintain a slightly better signal strength to maintain the same modulation rate. In practice, this will mean there is a slight trade-off with decreased modulation rate when increasing channel width, while maintaining all other variables constant (such as transmit power, antenna gain, etc).

To answer the questions surrounding the use of 256-QAM at distances greater than 10-20 feet, I've found that the usable distance and signal strength required to use 256-QAM data rates is around 25-30 feet (in my case with one light wall in-between) and around -52 dBm signal strength. I note multiple data rates being used for the 802.11ac client at Point #2 because the client appeared to be data rate shifting during the test, likely unable to sustain 256-QAM modulation at times and shifting to a lower rate. These values are almost certain to vary between client adapters based on receive sensitivity, but this should provide a rough estimate for WLAN administrators.

Results Recap

Let's add up the 802.11ac test results:
  1. Decent peak performance at 80 MHz channel width, although it we should see double this performance with integrated adapters or USB 3.0 external adapters.
  2. Better peak performance than 802.11n at comparable 40 MHz channel width due to the use of more aggressive 256-QAM at relatively close distances to the AP.
  3. Better rate-over-range performance than 802.11n, especially as distance from the AP increases and signal level deteriorates.

Final Thoughts
It was great to get a first-look at 802.11ac equipment, even though the currently available client adapters are a bit disappointing from a peak performance standpoint due to their reliance on USB 2.0 bus speed. You might consider waiting for integrated 802.11ac client adapters or external USB 3.0 adapters to hit the market, which should be capable of supporting the full throughput that 802.11ac offers. A few are already out there; I found this 802.11ac hardware wiki that seems to be keeping track of consumer equipment.

Even though you might not see awe-inspiring peak performance for any single client with this early release equipment, 802.11ac still stands to improve the aggregate performance and capacity of the network through more efficient use of airtime. WLAN administrators should expect to see this aggregate increase in network capacity even with 802.11ac capable mobile devices such as tablets and smartphones. Since they will be using higher Wi-Fi data rates, they will be getting on and off the air quicker for a given application throughput level than they would by using 802.11a/g/n. This will translate into the ability to support more clients or higher throughput per-client, and will be a big boost for enterprise WLAN capacity!

Cheers,
Andrew


Read the Entire 802.11ac Gigabit Wi-Fi Series:


10 comments:

  1. I've seen similar results with the Buffalo 11ac router while using a NETGEAR 11ac USB adapter. Sadly, that adapter is also USB 2.0. The interesting thing to me so far is how different the spectral representation is for varying devices. Your ASUS looks better than the Buffalo or the Linksys that I have tested. [I have the ASUS too, time to pop it out :)]

    It is also very interesting to watch the primary channel flow (typically from left to right with most setups) of energy. Particularly when you connect a few 20 MHz and 40 MHz devices at the same time.

    Great post as usual sir!

    Tom Carpenter

    ReplyDelete
  2. Andrew,

    I have been following your posts in regards to 802.11ac. They have been very informative to me and I appreciate all the data that you have shared.

    Nice simple to read. Thanks!

    EJ Jackson

    ReplyDelete
  3. Hi, Andrew

    We are working on something about latest WiFi hotspot models that can support multiple SSIDs, can you tell how many SSID can be configured into ASUS RT-AC66U?

    Thanks,

    Roy Fang from hash.ly

    ReplyDelete
    Replies
    1. Hi Roy,
      The ASUS router supports 1 SSID on 2.4 radio and 1 SSID on 5 radio which can be different. It also supports a separate guest SSID on each radio.

      Cheers,
      Andrew

      Delete
    2. Thanks, Andrew,

      Some Cisco models claim to support 16, maybe an overkill but definitely to our liking!

      Roy

      Delete
    3. Roy, please be careful to compare like with like. The Cisco model you are looking at is likely an enterprise AP while the Asus reviewed here is a home router/AP combo. Having 4 SSIDs available on a home router is quite a lot!

      Delete
    4. Andrew,

      Yes, I am sure they are in different price brackets as well.

      In fact, we are keen on number of SSID is based on our bold thinking about SSID broadcasting, that eventually can be a form of offline broadcasting of trending hashes up to 32 characters, and we have an Android app to capture it and connect to designated landing page, once internet connection is secured. Too complicated? well, just one of my science projects, so far.

      Yours,

      Roy

      Delete
  4. After all these years, you're still my favorite blogger. Dude, this series on 11ac has just ROCKED!

    Thanks for all of the hard work you've put into these blogs man.

    Devinator

    ReplyDelete
  5. Well done, and definitely answered a few questions I had about 802.11ac. Thanks!

    Ryan

    ReplyDelete
  6. I also have tested new 802.11ac APs and adapters. With about same slightly disappointing results.
    Still I doubt that you are hitting USB 2.0 speed limit. About double the transfer speed should be possible through USB 2.0 compared to what you got.

    Anyway, these are first 802.11ac devices, so I expect them to have lot of problems in their implementations. Looking forward for driver and firmware updates to these devices.

    ReplyDelete