Forum

Tony and I were discussing the video and beam forming in the roadshow blog and came to a discussion that I thought would be best in the forums since that is easier to keep up with vs. blog responses.

During the responses I said that 80% of traffic in the network comes from the AP, while Tony says it is more like 60-65%.

This number is important because of the ramifications that some technology has vs. others. Basically it comes to this: We can more easily control what the AP's can improve for performance much easier than the STA's. Networks with higher amounts of traffic from the AP can benefit more from technologies such as beam forming etc.

So, down to the numbers. The reality is, Tony is right in some cases and I am right in others. It just depends on the network. Let's take a look at a few examples:

- Surfing the Internet and checking email: 80-90% of traffic comes from the AP
- Total VoIP network only: 50/50 AP's vs STA's, assuming that each party on the endpoint of each call talks the same (good assumption).

Surfing and email shouldn't be much to debate. STA requests a web page (low byte count) AP delivers web page (much higher byte count)

Now, let's look at VoIP. Earlier I said it was 50/50 between STA's and the AP and that is correct. However, there is more to this story.

Now I want to make a chart to show something. In this example we are talking about VoIP STA's only.

- 1 AP - 1 STA - Each Tx's 50% of the traffic
- 1 AP - 2 STA - Now, the AP is transmitting more often than each individual STA.
- 1 AP - 10 STA's - If all 10 STA's are on a call, the AP will Tx 10 times for each of the STA's one Tx. Yes, if you look at all STA's combined it is still 50/50. However, each STA is an individual transmitter with different collision domains and different data rates.

The point here is this: With technology such as chip based (Cisco) and dynamic antenna (Ruckus) beam forming, we can increase the signal from the AP to the STA which improves data rate to each STA. There is currently no technology to do this from the STA side.

GT

I'm glad everyone has an opinion about this topic. Makes it fun. GT, thanks for being the bullseye here. :)

Regarding hidden nodes, several comments have been made that have highlighted the fact that a hidden "node" can be a client or an AP. GT's vlog was only intended to highlight the fact that there is such a thing as a hidden AP hidden node. Many folks think that hidden nodes are only clients. As he highlighted in the video, Ruckus' dynamic beamforming may help to eliminate some problems with hidden APs. 36bells (I believe) pointed out rightly that this type of dynamic beamforming may also cause hidden client nodes because clients may not know if the AP is transmitting in a different direction. This has been a traditional criticism of Ruckus' beamforming. This is always a tough topic without a whiteboard, but there are two important points here:

1. Ruckus' use of beamforming not only uses directional transmitters, but also directional receivers. So, you might get two clients that transmit at the same time, but a smart antenna array can reject a signal coming from one direction if that is not the intended direction of reception.

2. Let's say you have an AP and two clients, STA-1 and STA-2 on opposite sides of the BSS. If the AP transmits to STA-1 using a directional transmission and STA-2 (a hidden node) transmits at the same time (or shortly thereafter, b/c STA-2 didn't hear the APs transmission), STA-1s signal from the AP (a directional Tx) is likely to be much stronger than the signal from STA-2 (an omnidirectional Tx, usually). So, there may be a hidden node and a retry, but the benefit of high SNR between the AP and client prevents an adverse performance impact for STA-1s reception of the APs transmission.

Two more thoughts and I'll shut up.

GT makes a good point that, even if hidden node problems persist and you get more collisions, by using beamforming, many transmissions are performed at a higher data rate, which minimizes the airtime used for each Tx.

Also, I'd have to talk with engineers over at Ruckus, but I'm pretty sure that some transmissions use both directional and omnidirectional antennas at the same time. This would be helpful both to the destination station (higher SNR) as well as the surrounding non-destination STAs (can use standards-based contention).

Thoughts?

Interesting topic :)

I won't argue on the number 65 or 80% doesn't much of difference but I don't agree with your examples GT. The AP with DCF doesn't have more chances or more priority to access the air than the clients.
So what do you get if we take again your examples:

- 1 AP - 1 STA - Each Tx's 50% of the traffic
>> they have both 50% of chances to gain access to the medium that's different than saying that they have both 50% of the traffic that's simply not true.

- 1 AP - 2 STA - Now, the AP is transmitting more often than each individual STA.
>> not true the AP has 1 chance over 3 to gain access to the medium

- 1 AP - 10 STA's - If all 10 STA's are on a call, the AP will Tx 10 times for each of the STA's one Tx.
>> same thing the AP has 1 chance over 10 to gain access to the medium

Yes, if you look at all STA's combined it is still 50/50.
>> that's not true, there no way you can ensure this

Now comments on Marcus post

Point 1: that's true but you can't avoid a collision if ST'as are transmitting at the same rate. If not at the same rate then I agree.

Also, I'd have to talk with engineers over at Ruckus, but I'm pretty sure that some transmissions use both directional and omnidirectional antennas at the same time.
>>I don't think so it would be ineficient cause I'm not an antenna expert but I think you still have to stick to max EIRP

Roma,

Now you are getting into arbitration. :)

Let's use the example of 10 VoIP STA's connected to one AP. Just to be clear, we are saying that 10 STA's are all on a call. Your response is incorrect and missing one major point.

In your example, there are 11 devices that are actively sending VoIP data. Your math points to each device needing to Tx the same amount, and it just doesn't work that way. Each STA is transmitting data for it's user. The AP is transmitting data for ALL users. Given this example, the AP is transmitting 10 times as much data as any one STA. That is fact. :)

Here is another point that is often missed. Let's say you are on a VoIP call. How often do you think it is transmitting a data packet? Most people are under the impression that it is transmitting a frame every time it gets the chance and it just ain't so.

Surprisingly, there is a HUGE amount of space between each VoIP frame (we are still talking about one device here). Now, the AP is transmitting 10 times as much voice traffic than any one STA. So, it isn't contending with 10 devices at any one time, even if all of those STA's are all on a call. I know it feels weird.

I'll soapbox this for a moment. Here is how I can prove that 10 devices aren't arbitrating for the medium at one time:

In QoS, a VoIP device will randomize between 2^n-1 where n=2. This means that every VoIP device randomizes between 0-3 before transmitting. This is the contention window. If 10 devices were all contending, you are guaranteed collisions which wouldn't work. I mean it wouldn't work, not that it just wouldn't work well. Every frame would be a collision.

I guess I'm done with this post. More in a bit. :)

GT

Roma59 Escribi?3:

Interesting topic :)

Now comments on Marcus post

Point 1: that's true but you can't avoid a collision if ST'as are transmitting at the same rate. If not at the same rate then I agree.

Also, I'd have to talk with engineers over at Ruckus, but I'm pretty sure that some transmissions use both directional and omnidirectional antennas at the same time.
>>I don't think so it would be ineficient cause I'm not an antenna expert but I think you still have to stick to max EIRP

Point #1: What? You can't avoid a collision if they are at the same rate? Not sure where that is coming from.

A collision occurs when two transmissions are received by a device on the same channel. There is only a collision if the device can't distinguish between the correct signal and a secondary signal. For example, if two signals get to a STA at the same time, a STA can still interpret the signal IF there is enough SNR between the two signals. If one signal is received at -65 dBm and another at -80 dBm, there is a 15 dB SNR and could be properly received.

As far as your last comment, not sure about that. Max EIRP has nothing to do with how well a device can receive.

GT

GTHill Escribi?3:

Roma,


Here is another point that is often missed. Let's say you are on a VoIP call. How often do you think it is transmitting a data packet? Most people are under the impression that it is transmitting a frame every time it gets the chance and it just ain't so.

Surprisingly, there is a HUGE amount of space between each VoIP frame (we are still talking about one device here).
GT

Good point GT. A typical VoWiFi station with an active call has to transmit a frame maybe every 20 ms. Arbitration could occur in as short as 40 usec (or somewhere near that value) for a typical 802.11g phone. So, yeah, you're not usually talking about each STA competing for the medium right at the same time.

On point 1, can you elaborate Roma? As I understand your comment, I'm in agreement with GT on this one. Not sure what data rate has to do with the possibility of a collision. It's sort of like if you cup your ears to hear in front of you and someone behind you is talking... as long as the person in front of you is talking loudly enough and your ear cupping is done properly, you don't have to worry about hearing the guy behind you, even if he's speaking at the same pace (and volume) as the guy in front.

On the other point, we're talking indoor antennas and Tx power, so you shouldn't have issues with regulatory requirements for EIRP as it relates to omni- vs. semi-directional antennas. I'll readily admit that I may be completely wrong regarding how Ruckus does this.

Thanks for all the great comments and discussion! You guys are laying it on and I'm glad that you are. This is the kind of stuff that can either validate or invalidate all of our claims. I hope it is doing the former. Keep the harassment coming! :)

A collision occurs when two transmissions are received by a device on the same channel. There is only a collision if the device can't distinguish between the correct signal and a secondary signal. For example, if two signals get to a STA at the same time, a STA can still interpret the signal IF there is enough SNR between the two signals. If one signal is received at -65 dBm and another at -80 dBm, there is a 15 dB SNR and could be properly received

>> I'm not saying anything different just that english is not my mother language so sometimes it's difficult to find the rigth words to express my mind

As regards to EIRP what I meant is that the total radiated power can't exceed 100mw indoor, in other words if you transmit omni + directional you can'exceed 100mw in total, so either you decrease the omni side or the directional so It would be ineficient to use both at the same time. What Ruckus can do is for example send the becons using a directional pattern.

Ok sorry I missed your post on arbitration my favorite topic ;)

In your example, there are 11 devices that are actively sending VoIP data. Your math points to each device needing to Tx the same amount, and it just doesn't work that way. Each STA is transmitting data for it's user. The AP is transmitting data for ALL users. Given this example, the AP is transmitting 10 times as much data as any one STA. That is fact.
>> hmmm I won't disagree, but I wouldn't say "the AP is transmitting 10 times as much data as any one STA" rather say that it has to transmit 10 times more data (which is a little bit different) but doesn't mean that it will not have to compete against 10 STA's to get access to the medium

Here is another point that is often missed. Let's say you are on a VoIP call. How often do you think it is transmitting a data packet? Most people are under the impression that it is transmitting a frame every time it gets the chance and it just ain't so.
>> ok it's abvious if you take the specific case of of a VOIP device. It's not necessary true in the case of device sending data.

Surprisingly, there is a HUGE amount of space between each VoIP frame (we are still talking about one device here). Now, the AP is transmitting 10 times as much voice traffic than any one STA. So, it isn't contending with 10 devices at any one time, even if all of those STA's are all on a call. I know it feels weird.
>> well you take a specific example of 10 VOIP but still when 1 STA sent its packet, the other 9 STA's and AP are starting to contend...

In QoS, a VoIP device will randomize between 2^n-1 where n=2. This means that every VoIP device randomizes between 0-3 before transmitting. This is the contention window. If 10 devices were all contending, you are guaranteed collisions which wouldn't work. I mean it wouldn't work, not that it just wouldn't work well. Every frame would be a collision
>> hmmm ok, for each station you can specify cwmin/cwmax and to make it even more complex AIFS >= DIFS, so guess what happens when you have non WMM phones :)

If the phones are non-QoS capable they will of course randomize between 0-15 if OFDM or 0-31 if DSSS. However, that doesn't change my point. Even a device like a VoIP phone that will Tx a lot, doesn't actually transmit all of the time, even during a call.

Roma,

Thank you very much for your participation on the forums. I appreciate a good discussion as it helps everyone (including me) understand how these complicated devices actually work.

GT

Thx GT,

I think the point I was trying to make is that with WMM you can set CWMIN/CWMAX (contention window size Min / Max) per client dynamically so what you say can be rigth for your specific example but I can find 99 other cases where it won't work this way. So you can't conclude that Ruckus solves the hidden node issue.
As regards to AIFS >= DIFS, I won't continue on this cause it's not the topic but it has a wider impact then what we think when you have a mix of both wmm and non wmm clients.