Extremely long simulation time

[Mike Xue]

Hello,

I am running the "cttc-fh-compression.cc" file to generate some simulation results for a hexagonal grid deployment. I left most of the parameters to the default value,
the only things I modified are "numOuterRings = 3"; "ueNumPergNb = 25"; "scenario = "UMa"; and "simTimeMs = 500". This would generate a layout of 57 cells and 1425 UEs. On my machine, (intel i5-6500 @ 3.2GHz, 16 GB RAM), the simulation would take approximately 29 hours to complete. I also tried setting "traces" to "false" hoping to reduce the amount of time needed, but it doesn't seem to help.

Has anyone tried to run any simulations similar to what I am doing? How long would it take to run on your machine? and is there any modifications I could do to reduce the amount of time needed to run such a simulation?

Many Thanks,

happy holidays

Mike X.

[Sandra Lagén]

Hi Mike,

Yes, unfortunately, it is normal... Having a simulation with a lot of nodes iand the 3GPP spatial channel model, takes long time. What you can do is to reduce the number of antennas, to 1x1 for example, this would speed up significantly your simulations. Also, if you use a random traffic model, instead of CBR (used in that example) in which there are packets to all users in all the slots, the simulation time will be reduced.

BR,

Sandra Lagén

Missatge de Mike Xue <xuey...@gmail.com> del dia dj., 24 de des. 2020 a les 20:08:

--
You received this message because you are subscribed to the Google Groups "5G-LENA-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to 5g-lena-user...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/5g-lena-users/99095f4d-8702-49da-9762-8c33a9f38037n%40googlegroups.com.

[Amiraslan Haghrah]

Dear Sandra and Mike,

A while ago, I translated the 3GPP channel model which develop by Tommaso Zugno, to a GPU based code using OpenCL. It runs about x10000 faster specifically for large array antennas using NVIDIA GeForce RTX 2080 Ti graphic cards. But I didn't find research communities interest in this topic, So I didn't update it for new updates of the 3GPP channel model.

Also, the developed code runs inside ns-3 with some modification in the waf setting and doesn't need any communication with another environment.

Best regards,

Amir

[Zoraze Ali]

Hi Amir, 

Nice job!! I think your effort could be of interest to many people. I had a look at the code, and my understanding is that it is based on the old code. So, if you could port the latest code and can perform some benchmarking using CPU and GPU, it would be very valuable. 

Kind regards,

Zoraze

[Amiraslan Haghrah]

Hi, Zoraze

As you said it is based on the old version, I should update it with the new version. I had done this based on the first version of the 3GPP channel model developed by Tommaso Zugno at the University of Padova.

BR

Amir

[Mike Xue]

Hi Sandra,

Happy New year!

Thank you for your response, it was very helpful!

I have 3 quick follow up questions,

1. is there a "switch" that can turn on the random traffic model? or would I need to implement the model myself?

2. Part of my research topic also looks into the rural scenarios for the hexagonal grid layout. I see that in the file "hexagonal-grid-scenario-helper.cc", RMa scenario parameters are defined, However in the "cttc-fh-compression.cc" example that I am basing my work on, in the function "RadioNetworkParametersHelper::SetNetworkToNr " seems like only UMa and UMi scenarios are supported. I am a little confused, does the simulator support RMa scenario?

3. I am writing a MATLAB script to plot CDF curves for the SINR and Throughput values, based on the data from "SinrTrace.txt" and "DlRlcStats.txt". I noticed that the Cell ID in those files are all even numbers, for example I have 3 rings, so 57 cells, and they would be numbered 2 ,4, 6,...,114. But in the code "hexagonal-grid-scenario-helper.cc" when they are generated they are labeled as 0~56. I am wondering where did that transition happen, and how do the numbers map to each other?

Many thanks!

[Sandra Lagén]

Hi Mike,

Happy new year to you too! I include responses in line with your text. BR,

Sandra

Missatge de Mike Xue <xuey...@gmail.com> del dia dc., 13 de gen. 2021 a les 20:53:

Hi Sandra,

Happy New year!

Thank you for your response, it was very helpful!

I have 3 quick follow up questions,

1. is there a "switch" that can turn on the random traffic model? or would I need to implement the model myself?

SL: We have been working on FTP model 1, according to 3GPP, but this is not yet available, and will likely be included in a future release. This would be a good example for the randomization, with big files, with packet arrivals according to the FTP process. Another option is that you randomize the start time of the applications, and reduce the lambda value in the example (so that less packets are created per second, and so the offered load per UE would be lower, thus reducing the simulation time as well).

2. Part of my research topic also looks into the rural scenarios for the hexagonal grid layout. I see that in the file "hexagonal-grid-scenario-helper.cc", RMa scenario parameters are defined, However in the "cttc-fh-compression.cc" example that I am basing my work on, in the function "RadioNetworkParametersHelper::SetNetworkToNr " seems like only UMa and UMi scenarios are supported. I am a little confused, does the simulator support RMa scenario?

SL: Yes, we support RMa. You can find it defined inside "hexagonal-grid-scenario-helper.cc" (SetRmaParameters). The example you mention is parametrized for UMa and UMi only, so you could easily extend extend the example to be able to configure RMa deployment. Basically, include the power and the RMa for channel modeling, and remove the related asserts.

3. I am writing a MATLAB script to plot CDF curves for the SINR and Throughput values, based on the data from "SinrTrace.txt" and "DlRlcStats.txt". I noticed that the Cell ID in those files are all even numbers, for example I have 3 rings, so 57 cells, and they would be numbered 2 ,4, 6,...,114. But in the code "hexagonal-grid-scenario-helper.cc" when they are generated they are labeled as 0~56. I am wondering where did that transition happen, and how do the numbers map to each other?

SL: One thing is the cellId in the example (0,1,..56) to create the cells, the other is the cellId in the simulator. In the simulator, please have a look at  NrHelper::InstallSingleGnbDevice, the cellId depends on the number of bandwidth parts per node. Maybe you are using FDD or two BWPs per gNB? If so, yes, the cellIds are 2, 4, 6... if TDD and single BWP, I think that then they would be contiguous. At some point we have to change the way of labeling cellIds and bwpIds in the simulator, but we need some time for this...

 

Many thanks!

On Monday, December 28, 2020 at 8:16:15 AM UTC-5 amirasla...@gmail.com wrote:

Hi, Zoraze

As you said it is based on the old version, I should update it with the new version. I had done this based on the first version of the 3GPP channel model developed by Tommaso Zugno at the University of Padova.

BR

Amir

On Monday, December 28, 2020 at 2:07:23 PM UTC+3:30 zora...@gmail.com wrote:

Hi Amir, 

Nice job!! I think your effort could be of interest to many people. I had a look at the code, and my understanding is that it is based on the old code. So, if you could port the latest code and can perform some benchmarking using CPU and GPU, it would be very valuable. 

Kind regards,

Zoraze

--

You received this message because you are subscribed to the Google Groups "5G-LENA-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to 5g-lena-user...@googlegroups.com.

To view this discussion on the web, visit https://groups.google.com/d/msgid/5g-lena-users/6c93b3cb-d962-46e0-9701-82633f989055n%40googlegroups.com.

[Sandra Lagén]

Mike, regarding the CellId, it is documented in nr-module.rst (pasted below). So, your case is TDD with one BWP, but the labels are used to label the cellId and the bwpId, and for that reason they are not contiguous when looking only at the cellId. With the example below you will understand it better.

BR,

Sandra

The ccId numbering is, for some untrained eyes, weird. But that is because some numbers in the sequence are used to identify the Cell ID. For example, let's consider a scenario in which we split the spectrum into three parts. We have four GNBs, and the numbering will be the following:.. _tab-example-bwp:.. table:: CcId numbering example   +------------+------------+---------------+---------------+---------------+
   |    GNB     |  Cell ID   | CcId for BWP0 | CcId for BWP1 | CcId for BWP2 |
   +============+============+===============+===============+===============+
   |   GNB 0    |      1     |       2       |       3       |      4        |
   +------------+------------+---------------+---------------+---------------+
   |   GNB 1    |      5     |       6       |       7       |      8        |
   +------------+------------+---------------+---------------+---------------+
   |   GNB 2    |      9     |       10      |       11      |      12       |
   +------------+------------+---------------+---------------+---------------+
   |   GNB 3    |      13    |       14      |       15      |      16       |
   +------------+------------+---------------+---------------+---------------+
If we would use this as a simulation scenario, the messages that come from the CcId 2, 6, 10, 14, would refer to the same portion of the spectrum. These IDs, internally at the GNB, would be translated into the BWP 0 in all the cases. The BWP 1 will be associated with the CcId 3, 7, 11, 15 (respectively), and everything else comes easily.

Missatge de Sandra Lagén <sandra...@gmail.com> del dia dj., 14 de gen. 2021 a les 11:31:

[Mike Xue]

Hi Sandra,

This makes so much sense, Thank you for the help!

cheers

[Mike Xue]

Hello Sandra,

I tried out the hexagonal ring example with a some modification, and plotted the resulting UE SINR distribution. The values I got was higher than what I was expecting, I was wondering if you guys had done any similar simulations to compare the results with.

Here are the details:

I ran the "cttc-fh-compression.cc" example with the following configuration:

numOuterRings = 3;

ueNumPergNb = 25;

traces = true;

scenario = "UMa";

radioNetwork = "NR";

operationMode = "TDD";

udpPacketSize = 600;

lambda = 2000;

simTimeMs = 500;

udpAppStartTimeMs = 400;

direction = "DL";

as well as in "hexagonal-grid-scenario-helper.cc", in the function "HexagonalGridScenarioHelper::SetUMaParameters"

m_isd = 500;

Everything else was left as default. This gives me a layout of 19 sites (57 cells), and a total of 125 UEs, in the Urban Macro scenario with ISD 500m, and by the end of the simulation there will be 100ms worth of results.

For getting the SINR cdf graph, I used the output file "SinrTrace.txt" and calculated the average SINR for each UE located in the inner 7 sites (525 UEs in total), then plotted the CDF of the SINR values.

The SINR cdf plot is shown here (also in the attached file), If this is correct it seems that some UE are able to achieve SINR of 50, 60 or even 70 dB, which seems a little too high.

Has your group done any similar simulations? Does my result seem reasonable? Does it match what you might be getting?

I also attached the "SinrTrace.txt" output file and my MATLAB code for plotting if you are interested.

Many thanks!

[Mike Xue]

Hello,

The main issues with the SINR CDF plot are, the plateau around 40~60dB, and from the left most of the CDF curve seems like around 5% of the UEs are not getting any signal. Any thoughts on what might be causing it?

Cheers

[Sandra Lagén]

Hi Mike,

For the UEs not being served, this may be because of the access mode and scheduler under consideration. Can you confirm you use OFDMA and RR, and which is the bandwidth and RBG size considered in this example? Note that in case of TDMA and RR, then the number of UEs scheduled in a slot is limited to the number of symbols available for data in a downlink slot. Under TDMA, among the implemented schedulers, only PF scheduler keeps track of scheduled UEs across slots. In case of OFDMA, the number of UEs that may fit in a slot depends on the bandwidth and the RBG size. Please check this.

Regarding the percentage of UEs with high SINR values, this may be because there is not a minimum gNB-UE distance in the deployment (I am not sure about this now, if we considered a minimum distance or not, I would say we don't). Try to check it if you want. The deployment generator generates a gnuplot with the considered gNB/UE's positions.

Hope it helps.

BR,

Sandra

Missatge de Mike Xue <xuey...@gmail.com> del dia dj., 4 de febr. 2021 a les 4:39:

To view this discussion on the web, visit https://groups.google.com/d/msgid/5g-lena-users/7d04c3ae-ec0b-4e56-8ad5-bb6c8b686375n%40googlegroups.com.

[Mike Xue]

Hi Sandra,

Thank you for your response, I will check these!

cheers

[Mike Xue]

Hi Sandra,

When I ran the program, I didn't change any of the code related to the access mode and scheduler in the "cttc-fh-compression.cc" file, when I looked through it, I only see the scheduler set for the LTE case. In the function "Set5gLenaSimulatorParameters", i found the following code:

// Scheduler type
    if (radioNetwork == "LTE")
      {
        nrHelper->SetSchedulerTypeId (TypeId::LookupByName ("ns3::NrMacSchedulerOfdmaPF"));
        nrHelper->SetSchedulerAttribute ("DlCtrlSymbols", UintegerValue (1));
      }

However in the example I couldn't find where the scheduler is set for radioNetwork is "NR". Do you know what place this setting is being done?

Thanks so much

MIke X.

[Sandra Lagén]

Hi Mike,

Yes, sorry, in this version of the example, the default access mode and scheduler was being used (i.e., TDMA RR). So, you can use the lines below and test OFDMA-PF, TDMA-PF, and even OFDMA-RR could solve the issue of UEs not being served depending on the used bandwidth.

BR,

Sandra

  // Scheduler type
    if (radioNetwork == "LTE")
      {
        nrHelper->SetSchedulerTypeId (TypeId::LookupByName ("ns3::NrMacSchedulerOfdmaPF"));
        nrHelper->SetSchedulerAttribute ("DlCtrlSymbols", UintegerValue (1));
      }

    else  // NR

      {
        nrHelper->SetSchedulerTypeId (TypeId::LookupByName ("ns3::NrMacSchedulerOfdmaPF"));
        nrHelper->SetSchedulerAttribute ("DlCtrlSymbols", UintegerValue (1));
      }

Missatge de Mike Xue <xuey...@gmail.com> del dia dj., 11 de febr. 2021 a les 1:05:

To view this discussion on the web, visit https://groups.google.com/d/msgid/5g-lena-users/5b596484-2f3e-4a58-875f-3e2edfbc3c1fn%40googlegroups.com.

[Mike Xue]

Thank you Sandra! I will give it a try.

cheers