Rescom TP instructions

Instuctions for the Lab on Markov chain modeling and MDP analysis, at the RESCOM2019 summer school

Objective

The goals of the Lab session is

• program the model of a discrete-time queue with impatience, batch size and finite capacity, using the marmoteCore library;
• program the same model with a control of admission in service, with the marmoteMDP library;
• compute the optimal service policy in this queue.

Steps

Preparation

The first step is to have the library installed on your computer. Two possibilities:

• using a virtual machine with virtualbox
  • download VM + instructions from this page
  • copy it from USB drive, available at the conference.
• using the compiled library (linux only)
  • tarball + instructions from marmoteCore's site

The instructions with virtualbox are then:

• install the virtualbox software from its web site
• launch virtualbox
• click on "Machine > Add"
• enter the location of the virtual machine that has been downloaded
• select the VM (Rescom2019_TP) in the right-hand pane and click on "Start"
• log in with username/password pierre/Rescom2019*
• you should see a desktop with two folders: TP_Marmote and TP_MDP

Instructions for building Markov Chains

Testing the example provided

• Click on the TP_Marmote folder
• click on the file "example1.cpp" (or right-click then select "geany")
• a command-line terminal should appear at the bottom. Type

./example1

Example 1: construction of a discrete-time Markov chain on a 3-state space.

• the program takes as arguments:
  • n, a number of steps
  • p1 p2 p3, three probabilities summing up to 1, representing the initial distribution
• it outputs
  • the probability transition matrix
  • a trajectory x[0], x[1], ... x[n]

• run the example with values, e.g.

./example1 4 0.2 0.3 0.5

• use the editor to modify the code example1.cpp, in order to make state 2 absorbing
• compile by clicking "Construire > Make"
• execute again
• modify further the code to make it compute the value of the distribution after n steps:

 Distribution* trDis = c1->TransientDistributionDT( 0, n );
 trDis->Write( stdout, STANDARD_PRINT_MODE );

• compile and execute

Developing a new model

• download/copy the skeleton of the program from here.

The file contains:

• the "include" instructions necessary
• a "combinations" and "binomial" useful for computing some probabilities
• the template of a "MakeGenerator" function returning a matrix (type SparseMatrix)
• a main() function.