ns-3 wifi model validation versus Bianchi mathematical model

[Amr ABDELFATTAH]

Hi All,

i had done a simulation for a 802.11a using ns-3 to calculate the probability of collision for number of stations 5,10 and 20 under saturation condition (i assume that, when i used high offered load by each station i'm in saturation condition) i had found:

prob. of collision for 5 Sta = 0.237292679

 

prob. of collision for 10 Sta = 0.358245827

prob. of collision for 20 Sta = 0.466621401

Based on the bianchis' paper titled "Refinements on IEEE 802.11 Distributed Coordination Function Modeling Approaches" , the probability of collision should be 

prob. of collision for 5 Sta = 0.2919

 

prob. of collision for 10 Sta = 0.3934

prob. of collision for 20 Sta = 0.4846

i have two question:

1) How can i be sure that the queue of each station has a packet every time to transmit to fit the saturation condition ?

after analyzing bianchis' paper, he considered that if there is at least two packet collide together both will be considered as lost packet whereas it is not the case in ns-3 YansWifiPhy since it uses the interference helper to decide the first packet will be lost or not so 

2) How can i change the interference helper to discard the packet whenever collision happened not depending on interference energy?

Thanks    

[Sebastien Deronne]

Le jeudi 11 juin 2015 17:10:28 UTC+2, Amr ABDELFATTAH a écrit :

Hi All,

i had done a simulation for a 802.11a using ns-3 to calculate the probability of collision for number of stations 5,10 and 20 under saturation condition (i assume that, when i used high offered load by each station i'm in saturation condition) i had found:

prob. of collision for 5 Sta = 0.237292679

 

prob. of collision for 10 Sta = 0.358245827

prob. of collision for 20 Sta = 0.466621401

Based on the bianchis' paper titled "Refinements on IEEE 802.11 Distributed Coordination Function Modeling Approaches" , the probability of collision should be 

prob. of collision for 5 Sta = 0.2919

 

prob. of collision for 10 Sta = 0.3934

prob. of collision for 20 Sta = 0.4846

i have two question:

1) How can i be sure that the queue of each station has a packet every time to transmit to fit the saturation condition ?

By adding more packets in the queue than what can be handled by the wifi channel.
 

after analyzing bianchis' paper, he considered that if there is at least two packet collide together both will be considered as lost packet whereas it is not the case in ns-3 YansWifiPhy since it uses the interference helper to decide the first packet will be lost or not so 

2) How can i change the interference helper to discard the packet whenever collision happened not depending on interference energy?

Why don't you just put all the nodes at position 0.0,0.0,0.0 if you just want to compare with Bianchi model?
Bianchi model anyway does not consider distance between nodes.

 

Thanks    

[Konstantinos]

In addition to Sebastien's comments, would should also run the same scenario several independent times in order to provide statistical results
https://www.nsnam.org/docs/manual/html/random-variables.html#id1

[Amr ABDELFATTAH]

Hi Sebastien,

thanks for your reply. about my first question, i think i had done exactly what you asked me to do since i load each wifi station with high traffic to be sure that the queues is never empty but my question is how i can monitor the queue during the simulation to be sure that i am in the saturation condition to fit well Bianchi. About my second question, i had did exactly what you suppose but actually the problem is, how ns-3::Yanswifiphy works. As you see my probability of collision is better than Bianchi except for the 20 Sta. i think that because even there is a collision between some packets the Yanswifiphy still can receive the packet correctly depends on the utilization of ns-3::interferencehelper (Note: my collision probability has been calculated depends on the number of missed ack in Mac layer). so i need to whenever there is a collision, i should lose that packets and can never be correctly received by mac layer. 

  

thanks 

[Amr ABDELFATTAH]

Hi Konstantios, 

Thanks for your reply. Yes i had done the simulation several times with different seed also but still same results. 

[Sebastien Deronne]

1/An easy check is by adding traces.
You could add debug traces on EdcaTxopN in order to see if this gets printed:

          NS_LOG_DEBUG ("queue is empty");

2/ Then show me traces to prove this is really happening.
You think or suppose is not accurate, you need to investigate and check this is really happening.
Once again, check traces when setting all your nodes on default position (0.0,0.0,0.0).

[nich....@gmail.com]

Hi,

have you completed the validation of bianchi model? I just want to compared the throughput using ns-3 simulation with the result calculated by bianchi model, but the result is not good. What about yours ?

在 2015年6月18日星期四 UTC+8下午5:00:59，Amr ABDELFATTAH写道：

[Zoraze Ali]

Hi Amr,

I guess you have already searched among the old discussions for similar topic but just in case you didn't, following discussion might help. 

https://groups.google.com/forum/#!searchin/ns-3-users/probability$20of$20successful$20channel$20access/ns-3-users/7byjPHGIcmo/f7UivGZKj5cJ

regards,
Zoraze

[Amr ABDELFATTAH]

Hi, 

I have some results which i like to share it with you and have your comments on it. I had used the equations of bianchi's' paper titled 'Refinements on IEEE 802.11 Distributed Coordination Function Modeling Approaches'. More precisely, equations (15) through (24) for basic access (without RTS/CTS).

And ns3 simulation parameters was

1/ channel bit rate = 6 Mbps.

2/ packet size = 512 byte at application layer using ON/OFF application thats is means (512+8+20+36) bytes at Mac layer.  (Bianchi uses 1500 byte at PHY layer)

3 / the throughput has been evaluated at the application layer in constract to bianchi throughput which calculated just before the MAC layer. 

So, Normaly My figures should be the same like Binachi's figure at page 1064 for the probability of collision since it does not depend on the packet size whereas for the throughput it is different (I had customise its equation of thoughput to meet my simulation results). 

 

[Kang Hailong]

Hi，Amr

The results seem right with the model.

Could you share the code?

The ns version is ns-3.23 or 3.24?

How did you customise the equation?

thanks

kang

在 2015年12月22日星期二 UTC+8上午12:33:39，Amr ABDELFATTAH写道：

[Amr ABDELFATTAH]

Hi,  

i had did a mistake in the equation of bianchi. the new figure for the probability of collision is better. i don't have any problem to share the code and i had used ns-3.24. about the equation, In bianchi's' paper the throughput is calculated at the network layer as it seems for me (that is mean the payload includes both the network and the transport layer header). so in my figure i show the throughput in application layer not like him (figure 6 page 1064). that is explain why my throughput is less than his throughput.

Thanks

Amr

--
Posting to this group should follow these guidelines https://www.nsnam.org/wiki/Ns-3-users-guidelines-for-posting
---
You received this message because you are subscribed to a topic in the Google Groups "ns-3-users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/ns-3-users/Plz4Q4Q2pAc/unsubscribe.
To unsubscribe from this group and all its topics, send an email to ns-3-users+...@googlegroups.com.
To post to this group, send email to ns-3-...@googlegroups.com.
Visit this group at https://groups.google.com/group/ns-3-users.
For more options, visit https://groups.google.com/d/optout.

[Amr ABDELFATTAH]

Hi, 

My comments on the figures.

1/ the collision probability shows the collision probability of bianchi is close to that of ns-3 for large number of stations , Which is normal due to bianchis' assumptions (as the number of station increases, as more the precise of his model get better).

2/ the throughput: 

 - first part: where ns3 throughput overestimate the throughput is normal due to the inaccurate measuring of the probability of collision. 

- second part : where bianchis' throughput overestimate the ns-3 throughput whereas the probability of collision are same. that is beacause the markov chain of bianchis' has a some false transition probabilities (the fact, he assumed that the unfinite retransmission limit). refereneces 

http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=983630&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D983630

Thanks 

Amr

[kang hailong]

hi，amr

can you share the code?

kang

--

[Amr ABDELFATTAH]

Hi,

yes, i will soon but as my code i had done since a long time since the date of that discussion 11/6/2015, i should to clean it :) , for all the persons can understand it.

best regardes,

Amr    

[Amr ABDELFATTAH]

Hi, 

following my last message, i had applied the model proposed in http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=983630&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D983630   as a correction for bianchis' model.  I think that the results better globaly for both probability of collision and throughput. what do you ? 

Best regards

[Amr ABDELFATTAH]

[kang hailong]

Hi,Amr

The results are good, but still not consistent with model. I think the topology and parameters have a big effect on the results.

I followed one topology in the 3 page at the website http://mobnet.epfl.ch/slides/C2-80211-PerfSlides.pdf.

But the results are not correct.

what about your topology?

Kang 

[Amr ABDELFATTAH]

Hi Kang,

My topology is exact like the right topology in the 3 page and more that all the sta nodes and Ap are position in the origion.

[Tommaso Pecorella]

Hi,

sorry to point out this, but if you want to validate a model with the simulations, you'll need to add some error bars and consider the standard deviation of the model and the simulations. The fact that the two curves are close enough and they're both (apparently) following the same trend isn't really a proof.

Add the standard deviation and/or do some box plots around each point. Perhaps you'll find that the curves are well within the statistical limits.

Cheers,

T.

[Amr ABDELFATTAH]

Hi, 

Thank you for your comments and i am completely agree with you. I hope soon i can repeat the simulation more times to do that statistics. 

Best regards 

Amr 

[Amr ABDELFATTAH]

Hi All,

again concerning Bianchi Vs NS3, for different packet size.

[kang hailong]

Hi,Amr

it looks good! 

But I am not sure that the relationship between number of frame and packet size can be analyzed in the Bianchi model.

The Application layer is responsible for generating the packet. So how did you set the simulation parameters?

kang

best regards

2016-04-03 23:39 GMT+08:00 Amr ABDELFATTAH <amr.s.ab...@gmail.com>:

Hi All,

again concerning Bianchi Vs NS3, for different packet size.

[Amr ABDELFATTAH]

Hi kang,

you can found the number of sended frames , collided frames and sucessfull frames using the equation of the throughput of bianchi and the probability of collision.

At the application layer you need to have only a saturation means all the time you have a packet to send (queue is not empty).

best regards 

Amr

[Amr ABDELFATTAH]

Hi All,

For my research i need to know the number of backoff slot before any transmission in Wi-Fi network whether it is a success or a collision. if someone has a idea how i can capture this information from the simulator it will help me. Otherwise  i had made a Matlab simulator for Wi-Fi to get that information. but still need to verified with Ns3. i show a comparison between Matlab simulator and Ns3 also for number of station Wi-Fi = 10.  

[srijan shakya]

Can you please upload the i am having some problem in it.