To reproduce simulation results (Figure 9-13) of MQ-ECN paper, you need following softwares:

• Network Simulator (NS) 2.35
• mqecn.patch (Modify TCP & Add ECMP)
• MQ-ECN code (implement DWRR and WRR with different ECN marking schemes)
• Simulation scripts (run simulations in parallel and parse results)

Download Network Simulator (NS) 2.35 and unzip it.

$ tar -zxvf ns-allinone-2.35.tar.gz

Copy mqecn.patch to the top ns-2.35 folder (ns-allinone-2.35) and apply the patch. Then install NS2.

$ cd ns-allinone-2.35
$ patch -p1 --ignore-whitespace -i mqecn.patch
$ ./install

Copy files of MQ-ECN code to ns-allinone-2.35/ns-2.35/queue/

Add queue/wrr.o queue/dwrr.o to ns-allinone-2.35/ns-2.35/Makefile

Run make on /ns-allinone-2.35/ns-2.35

Before running large-scale simulations, you can run several small-scale simulations to understand the basic properties of MQ-ECN. The scripts for basic simulations are test-wrr.tcl, test-wrr-2.tcl, test-dwrr.tcl and test-dwrr-2.tcl. The basic simulation scripts will generate three files: throughput.tr, qlenfile.tr and tot_qlenfile.tr.

• The throughput.tr gives aggregate goodputs of different services. It has the following format:

  Time, Goodput of Service 1, ..., Goodput of Service N
  

• The qlenfile.tr gives queue lengths (in bytes) of different service queues. It has the following format:

  Time, Queue Length of Service Queue 1, ..., Queue Length of Service N
  

• The tot_qlenfile.tr gives the total queue length (in bytes) of the congested switch port. It has the following format:

  Time, Queue Length of Switch Port
  

At the beginning, we briefly introduce the usage of each file in scripts directory.

• CDF_cache.tcl, CDF_dctcp.tcl, CDF_hadoop.tcl and CDF_vl2.tcl give flow size distributions used in our paper.

• spine_empirical_diffserv.tcl and [tcp-common-opt-raw.tcl] (https://github.com/HKUST-SING/MQ-ECN-NS2/blob/master/scripts/tcp-common-opt-raw.tcl) are NS2 TCL simulation scripts.

• run_simulation_diffserv.py is used to run NS2 simulations in parallel.

• result.py is used to parse final results.

Hence, to run simulations in parallel:

python run_simulation_diffserv.py

There are many parameters to configue in run_simulation_diffserv.py. Note that you need to modify ns_path and sim_script correspondingly. By default, it runs 20x simulations in parallel and iterates over the per-queue ECN with the standard threshold, per-queue ECN with the minimum threshold and MQ-ECN. For each simulation, it will create a directory with the following name:

diffserv_[scheduler]_[transport]_scheme_[ECN scheme]_load_[network load]_service_[number of queues]

Each direcotry consists of two files: flow.tr and logFile.tr. The flow.tr gives flow completion time results with the following format:

number of packets, flow completion time, number of timeouts, src ID, dst ID, service (queue) ID

You can use result.py to parse flow.tr files as follows:

$ python result.py -a -i [directory_path]/flow.tr

If you have any question about MQ-ECN simulation code, please contact Wei Bai.

We thank Mohammad Alizadeh for sharing pFabric simulation code and DCTCP patch.