Simulation

Introduction

Problem formulation has been implemented in IBM CPLEX. This formulation can be configured in many aspects and reflects the VNFM placement problem with VNFM in a data center. Results are placed in the results folder along with their configuration.

Up and Running

For running simulation, you need a working installation of CPLEX and Java/Maven on your system then you can run the simulation with the following command:

cd simulation
CPLEX_LIB="$HOME/ibm/ILOG/CPLEX_Studio128/cplex/bin/x86-64_linux"
test -f cplex.jar || echo "Please provide 'cplex.jar' in the current directory"
gradle build
sed -i "s#cplexLibrary =.*#cplexLibrary = ${CPLEX_LIB}#g" build.gradle
gradle run --args config

Exact Results

There are the reults of using this simulation for solving the real problem of placing the NFV chains on k-ary fat tree. In these results there is no constraint on license fee of VNFMs.

4 vCPU, 16GB of RAM, without GC limit, 12GB heap size, 8 chains

4 vCPU, 16GB of RAM, without GC limit, 12GB heap size, 12-ary fat tree

8 vCPU, 22GB of RAM, without GC limit, 12GB heap size, 12-ary fat tree

Heuristic Results

Each instance has a price between 20 to 30. We have 100 chains, and each chain has a length between 5 to 7. I have run each test with 3 different versions of the algorithm and report the proportion of the final revenue to optimal revenue.

As the numbers show, the placement algorithm do better when the per instances revenue increase.