Does the slowest AC client slow down the rest?

[chipped]

Hey guys,

I have a Wave 1 AC router, the Asus RT-AC3200 running Merlin firmware.

I have observed that on one of the 5GHz (5G2) radios I can only achieve 20MHz, whereas on the other (5G1) I can achieve 40MHz.

The difference is I only connect my MacBook Pro 3x3 and Apple TV 4th Gen and 4K to my (5G1) because I want to keep it fast.

I'm assuming that because the Apple TV only do 40MHz its not allowing my MacBook to establish a 80MHz link on 5G1?

And on 5G2 it must have devices that are only capable of 20MHz AC so it slows the rest down.

Which leads me to my final question, does the slowest client limit the rest?

Does Wave 2 AC address this?

 

[thiggins]

Make sure you are not confusing number of streams (transmit/receive chains) and bandwidth.
Regardless of # of chains, 802.11ac devices can support 20, 40 or 80 MHz bandwidth on 5 GHz.
Each bandwidth has its maximum link rate, which is 433 Mbps for single stream, 866 for two stream, 1300 for three and 1733 for four.

Devices should each be able to connect at their maximum rates when they are close to the router. 802.11ac also has the ability to dynamically change bandwidth on a frame by frame basis.

So, no, your slower AC devices should not slow your faster ones.

 

[chipped]

Hi thiggins, yes I'm aware of of streams and channel width.

While clients can connect at their maximum rates, what I'm asking is if one device connects at a channel width of 20MHz will that limit the rest on that radio to 20MHz also?

 

[thiggins]

Hi thiggins, yes I'm aware of of streams and channel width.

While clients can connect at their maximum rates, what I'm asking is if one device connects at a channel width of 20MHz will that limit the rest on that radio to 20MHz also?

I answered that. No. Bandwidth is dynamically adjusted for each STA.

 

[chipped]

Ok, I think the issue is being caused by channel availability and Asus not letting us select country codes. I can see when I scan my area there is a 5GHz network advertising 80MHz channel width (AU country code), however my RT-AC3200 will only broadcast 20MHz on 5G2 and 40MHz on 5G1 regardless of the settings I adjust.

I suspect this is because the AU country code allows more channels and the router decides its happy to advertise a 80MHz connection because there are enough non overlapping channels.

 

[sfx2000]

I answered that. No. Bandwidth is dynamically adjusted for each STA.

To extend and elaborate - 11ac can adjust BW and MCS per client and conditions - it's one of the real benefits of 11ac over 11n and earlier tech.

 

[Pinho2288]

Even though 802.11ac can adjust those parameters individually for each client the stations with slower link rates are gonna suck up more airtime than those ones with higher link rate, right? I would love some recommendations of good books or articles that go really in depth about how wireless works If you guys could give me any.

 

[thiggins]

Even though 802.11ac can adjust those parameters individually for each client the stations with slower link rates are gonna suck up more airtime than those ones with higher link rate, right? I would love some recommendations of good books or articles that go really in depth about how wireless works If you guys could give me any.

Matthew Gast's 802.11 Wireless Networks and 802.11ac A Survival Guide are my go-to's.

 

[GromitD90]

I think you will find that dynamic bandwidth adjustment in 802.11ac is not a widely supported feature.

As Matthew Gast's 802.11ac Survival Guide explains. Prior to sending a data packet at a reduced bandwidth the AP must send a special RTS packet with the group bit set in its mac address (also known as RTS w BW support). All 802.11ac devices are required to support receipt of these RTs's and respond accordingly with the appropriate CTS's.

Only since the introduction of Wave 2 devices have any AP's supported sending RTS w BW and few seem to do so. If you look on the WiFI Alliance website and look at the certificates for Wave 2 devices RTS w BW is one of the tested features (never was with Wave 1 devices). Unfortunately you can not search their database with a filter for that feature so the only way is to download each certificate and read it.

It is difficult to determine if an AP that does not have a WiFi certificate actually support RTS w BW since devices do not exchange this information in Beacon messages or during association. And vendors never seem to mention it in their literature.

I always thought dynamic bandwidth was in theory a nice feature so when I found a consumer class AP that supported it, according to its WiFI Alliance Certificate, I bought one. It is a Linksys EA6350v3. I ran a number of tests where the Linksys was set for 80MHz operation and a neighbouring 802.11n router was set up on one of the 20MHz channels in the secondary 40MHz of the Linksys's 80MHz. I reasoned that if I had a lot of traffic to and from an 802.11ac client associated with the linksys router and then introduced heavy traffic on the 802.11n channel - then the ac client and router (at least for the traffic from the AP to the client) should drop down to 40MHz operation. I tried monitoring what was going on with a spectrum analyzer and saw no evidence supporting that. So I did a number of packet captures and found for the Linksys AP:

1. Normal mode of operation is not to use RTS/CTS before sending data frames. Presumably, if CCA shows all 80MHz free it just transmits an A-MPDU.
2. When it does send an RTS it is always a bandwidth sensing RTS.
3. It only appears to send an RTS if it does not receive an Ack to a previous A-MPDU. In this event it sends an RTS w BW signaling before sending a retry frame.
4. Since I cannot capture the non-HT duplicates frames on the non-Pri channels I have no way of knowing if it is attempting to use the full 80MHz bandwidth when it does send an RTS.
5. I have tried captures on the non-Pri 20MHz channels and do see evidence of RTS/CTS activity.
I reran the tests using a Netgear R7000 (Wave 1) in lieu of the Linksys AP and saw no evidence of any dynamic bandwidth exchanges.

If you look at the reference model AP's submitted for WiFi Alliance certification from the chip vendors. I don't see any evidence that Broadcom Wave 2 chips support RTS w BW but those from Qualcomm and Mediatek do. A friend ran a more sophisticated test than mine to look for evidence of dynamic bandwidth usage using an Aruba 325 AP which is a pretty serious piece of equipment (not consumer class). We couldn't find any evidence and BTW the Aruba 325 uses the Broadcom chipsets.

So based on my readings and findings I would claim that no Wave 1 AP's support dynamic bandwidth and support with Wave 2 consumer class AP's is very limited both in ability to support and usage of the feature when it it supported.

I realize I have limited data points but I would welcome any feedback that would prove me wrong. I would be happy to share more details on my testing including packet captures.
Mike

 

[thiggins]

Thanks for sharing your analysis, Mike. Like many things Wi-Fi, spec is one thing, implementation is another.

I haven't done any mixed device testing since N and G days. Might be time for a quick article.

In busy WLANs, it remains good practice to separate slower and faster devices on to different radios where practical. Unfortunately, this must be done manually. Tri-radio routers with "Smart Connect" don't do a very good job.

Did you do any poking at Airtime Fairness implementations? That's another black box some consumer routers are using now.

 

[GromitD90]

To be honest Tim I haven't investigated Airtime Fairness implementations maybe I might give it a go sometime. However from the tests I ran I did see some interesting behaviour which you may be interested in. The tests basically consisted of a PC associated with the Linksys 802.11ac AP and an ipad Air associated with the 802.11n AP. The only device that was moved during the tests was the iPad. The PC was running a LANSpeed Test to generate traffic and the ipad was running ookla speed test. The initial goal was to investigate dynamic bandwidth support as described above but the speed test results were also interesting.

Running the Ookla speed test on the iPad with no 802.11ac activity I got around 80Mb/sec download and 11Mb/sec upload. Running the speed test while there was a load between the 802.11ac STA’s the download speed dropped dramatically (averaged from 0.5Mb/sec – 3Mb/sec) unless the iPad was fairly close to the 802.11n AP. Upload was only slightly affected dropping down to about 9Mb/sec.

Why did my 802.11n BSS exhibit such abysmal performance when running the download portion of the speed test while performing quite well in the upload portion? I believe it is due to 2 factors the primary one being the difference in CCA for 802.11n devices and CCA for 802.11ac devices, and the second factor being the use of RTS/CTS (not BW sensing just regular old RTS/CTS).

Using an iPad Air I measured the signal strength RSSI of the devices relative to one another (eg., what was the RSSI of the 802.11n AP as measured at the physical location of the 802.11ac AP). I realize that the RSSI numbers are relative and that taking readings with a different device I can get different results but they were a good enough approximation.

RSSI numbers were:
At the 802.11ac location RSSI of the 802.11n AP was -80dbm
At the PC location associated with the 802.11ac AP the RSSI of the 802.11ac AP was -33dbm and the RSSI of the 802.11n AP was -77dbm
At the location of the 802.11n AP the RSSI of the 802.11ac AP was -72dbm

The RSSI of the AP's measured at the ipad air location varied as I re-ran the test with it at various locations.
I think it’s likely that the 802.11n router when sending data frames to its client (download part of test) frequently sees that the 20MHz channel is clear when performing CCA (since measured signal level was around the threshold level of -82dbM) and so transmits the frame. The 802.11ac router and client however have a lower threshold on CCA for that channel (-69dbM) so it’s highly likely that they will see the channel as clear and also transmit frames – resulting in many collisions and retries (hence poor download numbers).
However, for upload (iPad to 802.11n router) the iPad, when performing CCA, sees much stronger signals from the 802.11ac STA’s and coupled with the iPad prodigious use of RTS/CTS (non BW sensing) the OBSS’s behave well together and the 802.11n BSS gets its fair share of airtime.

I have some other findings on the use of CTS to self by AP's that you may also find interesting.

 

[thiggins]

Wouldn't airtime fairness, i.e. transmit queue management, apply only to devices associated to the same AP? At any rate, there is no standard I know of for this feature. So anything goes!