Data Structures

BCP is written to allow for running in parallel. The serial version emulates a parallel environment. Because of this, every data structure that is passed between processes must be packed into a buffer. For most built-in types, BCP_buffer has an overloaded pack and unpack routine. For every user-defined class, we must write our own pack and unpack routine. Typically, these consist of a collection of calls to BCP_buffer::(un)pack() on each data member of the class.

• Member/AAP.cpp
  The AAP class simply defines a storage container for an instance of AAP. This includes the dimension of the problem $n$ and the costs $c_{ijk}$. It also includes some helper functions for indexing and debugging.

• include/AAP_user_data.hpp
  This class, AAP_user_data, is derived from BCP_user_data and is meant to store extra information that is to be sent between nodes after branching. In our case, since we branch on the original variables $x_{ijk}$ and not the master variables $\lambda_s$, we need to keep track of which variables have been fixed to 0 or $1$ at a particular node. We need this information so that we can enforce these conditions in our column generation subproblem.

• include/AAP_var.hpp
  Each variable (a member of ${\cal F}'$), represents an assignment between sets $J$ and $K$. This class, AAP_var, is derived from BCP_var_algo, which is a generic class for algorithmic variables. We use algorithmic variables (as opposed to indexed variables) since there are potentially too many to give an index to each assignment point. The storage of an assignment is done using a vector of the $(i,j,k)$ indices and its cost.