For this homework you need to login to one of the hosts in the
DECF linux clusters: http://www.decf.berkeley.edu
I recommend 'putty' for secure remote login from any computer
connected to the internet: http://www.putty.org/
Alternatively, you may also use 'ssh'.
If you have not used unix/linux before, here is a tutorial
explaining the basic unix/linux commands:
http://www.ee.surrey.ac.uk/Teaching/Unix/
If you do not like this one, google. Many more are available
on the internet.
For this homework you do not need to know much beyond the basic
unix/linux commands that are needed to navigate in the file system
and create/copy/delete files/directories, such as:
passwd (to change your password)
mkdir
cd
ls
cp file1 file2
mv file1 file2
more file2
rm file2
IMPORTANT: You must login using your course username. If you
login with another username, you cannot run cplex or ampl.
Login first to decf.berkeley.edu. This will put you to one
of the decf servers. Do not run cplex or ampl on the servers.
Instead, ssh to one of the hosts listed at
http://www.decf.berkeley.edu/ganglia/
to run your experiments.
___________________________________________
IEOR264 - HW #2 - Due: 02/20/08
1. Experiment with the two different models given in class for each of
the uncapacitated facility location and uncapacitated lot-sizing problems.
The objective of this homework is to compare the ease of solvability of
different MIP formulations and the impact of cplex options/modules on the
solution process.
IMPORTANT: Start early. If you leave it to the last day, you will not
be able to complete the experiments.
Follow the Sample Files link at the course web page
http://ieor.berkeley.edu/~atamturk/ieor264/
and download ufl.* in the directory samples/ampl to get started.
Run the following command
% ampl < ufl.ampl
to create the ufl.mps input file for cplex.
Read ulf.mod, ufl.dat, and ufl.ampl to see what ampl does.
Then read ulf.cpx to see the cplex options set in this file. You will
see that cplex cuts, preprocessing, heuristics are turned off. Node
selection is set to depth-first search, variable selection is set to
maximum integer infeasibility. This setting is close to the vanilla
branch-and-bound algorithm.
Run the following command
% cplex < ulf.cpx
to solve ulf.mps.
At any time you can hit Control-C to stop running cplex.
You will do six experiments for each model by turning on cuts,
preprocessing, heuristics, and changing the node selection rule to
best bound and the variable selection rule to strong branching one
at a time.
Compare the number of variables and constraints, the LP relaxation bound,
the time to solve the LP relaxation, the number of nodes in the branch-and-
bound tree, the overall time required to solve for the models. For the
LP stats, you need to solve the LP relaxation of the models by relaxing
integrality using the command 'change problem lp' in cplex.
IMPORTANT: Limit your computations to max 30 mins and 100MB b&b tree
memory. Otherwise, DECF staff will kill your runs.
Report your findings in a table and discuss them in a few paragraphs.
What is your conclusion from the experiments? If you just report tables
without a discussion explaining your observations in the tables and
conclusion, you will not get full credit.
Based on these examples, create the files for the uncapacitated
lot-sizing problem and repeat the experiments for the lot-sizing
problem.
Cplex has literaly hundreds of options to control the solution process.
The ones listed above are the most basic ones.
To see the available options and values run the following:
% cplex
% cplex> set
% cplex> Parameter to set: mip
% cplex> MIP parameter to set:
The rest of self evident.
To learn the complete set of cplex options, see the manual at
http://www.decf.berkeley.edu/help/apps/ampl/cplex-doc
The users' manual is hundreds of pages, so please do not print it.
You can send the cplex output to a file by typing
%cplex < ulf.cpx > ulf.out
If you have questions send Vishnu or me an email. Enjoy!