Originally posted by: DSL

Dear all,

I am currently wondering how to correctly understand CPLEX's log file produced when solving a transportation problem using netopt.

Here is the log file:

-------------------------------
Tried aggregator 1 time.
No LP presolve or aggregator reductions.
Presolve time = 3.82 sec. (1031.39 ticks)
Extracted network with 4000 nodes and 4000000 arcs.
Extraction time = 1.42 sec. (234.69 ticks)
Iteration log . . .
Iteration:     0   Infeasibility     =        592888.000000 (0)
Iteration: 10000   Infeasibility     =           926.000000 (295981)

Network - Optimal:  Objective =   2.9649600000e+005
Network time = 1.09 sec. (192.88 ticks)  Iterations = 10789 (10222)

Dual simplex - Optimal:  Objective = 2.9649600000e+005
Solution time =    6.86 sec.  Iterations = 0 (0)
Deterministic time = 1718.68 ticks  (250.39 ticks/sec)
--------------------------------

1) Since I'm interested in the total time CPLEX expends for finding an optimal solution to the network flow problem: What is the difference between "Network time" and "Solution time"?

2) Regarding the antepenultimate line: Am I misinterpreting the log or does it really mean that CPLEX invokes a dual simplex? (Why?)

Thanks for your comments!

Originally posted by: RWunderling

When applying the network optimizer to a problem given as an LP CPLEX performs the following steps:

1. Extract an embedded network from the LP and instantiate a network problem (with nodes and arcs)
    You'll find this step reflected in the two lines starting with "Extract..."

2. Solve the network problem and get the optimal basis
    You'll find this step in the following log lines until "Network time =..."

3. Install the optimal basis to the LP (potentially extending it if needed)

4. Solve the LP from this advanced basis
    You'll find this step in the remaing log lines.

If the full LP problem is a network, the 4th step is not going to produce any iterations.  However, all LP Simplex data structures need to be setup.  In particular, this includes the factorization of the basis matrix and the computation of the primal and dual solution vectors using said LU factorization.
 

Note, that using the C API, you can directly formulate your problem in terms of nodes and arcs using the CPXNET* functions.  This way you will only deal with step 2 described above, since 1, 3 and 4 are no longer needed.

Originally posted by: DSL

Thanks R. for giving a precise explanation again!