This folder contains 
Mixed Integer Programming Models (MIPs) 
from Capacitated Network Design problems in *.mps format.
The formulations are based on models proposed by SNDlib:
http://sndlib.zib.de/home.action

They were used for calculations for the INOC 2007 conference.
A first version of the conference paper has been published here:
http://www.optimization-online.org/DB_HTML/2006/11/1525.html

You are invited to use these instances for your research and computations 
and to compare your lower and upper bounds with those that we published in SNDlib:
http://sndlib.zib.de/home.action

!!!!!!!!!!
But be aware of the following:

When I was implementing my code, I was interpreting routing cost for bidirected
problems (bidirected link model) in a different way than SNDlib.

Let $e=ij$ be a link of the network having routing cost c_e and
let $f_ij, f_ji$ be the flow on $e$ in both directions
my interpretation:
the total routing cost of $e$ is  $c_e( f_ij + f_ji)$ 
SNDlib interpretation:
the total routing cost of $e$ is  $c_e( max(f_ij, f_ji))$

So, if you use the corresponding MIPs from this folder 
your calculations are done with a different objective than proposed by SNDlib.
There are 5 such instances (bidirected link model and routing cost):
atlanta-D-B-M-N-C-A-N-N
cost266-D-B-M-N-C-A-N-N
cost266-D-B-E-N-C-A-N-N
newyork-D-B-M-N-C-A-N-N
newyork-D-B-E-N-C-A-N-N

For these problems you cannot compare any lower or upper bounds with 
those in SNDlib. We only published upper bounds from (still feasible)
solutions but we transformed them to the correct (SNDlib-correct) objectives.

Here are lower, upper bounds, gaps and computation time 
with my interpretation of routing cost (without any warranty):
atlanta-D-B-M-N-C-A-N-N:
46244642.400    46244642.400    0.0%   1.1s
cost266-D-B-M-N-C-A-N-N:
18443128.294    18975081.600    2,9%   3600s
cost266-D-B-E-N-C-A-N-N:
18880614.264    19877393.040    5.3%   3600s
newyork-D-B-M-N-C-A-N-N:
886506.092      999702.000      12.8%  3600s
newyork-D-B-E-N-C-A-N-N:
888020.441      1009702.000     13.7%  3600s


Nevertheless,
  Happy computing

Christian Raack