README

Go to the documentation of this file.

CURRENT STATUS OF DEVELOPMENT:
 
UG framework, ParaSCIP(ug[SCIP,MPI]) and FiberSCIP (ug[SCIP,Pthreads]) are available as a beta version.
For MIP solving, ParaSCIP and FiberSCIP are well debugged and should be stable. For MINLP solving,
they are relatively stable, but not as thoroughly debugged. This release version should handle
branch-and-cut approaches where subproblems are defined by variable bounds and also by constrains
for ug[SCIP,*](ParaSCIP and FiberSCIP). Therefore, problem classes other than MIP or MINLP can be handled,
but they have not been tested yet.
 
INSTALLATION:
 
This is a very rudimentary description of what's needed to get FiberSCIP and ParaSCIP running.
 
1. You need SCIP version 3.0.0 or higher and a reasonable compiler suite with
   libraries installed.
   Under Ubuntu, you can install these with
     "sudo apt-get install build-essential libreadline-dev libz-dev"
   With OpenSuSE 11.4and 12.1 this is
     "sudo zypper in -t pattern devel_C_C++ && sudo zypper in readline-devel zlib-devel-static"
   The package names may differ on your linux distribution.
 
2. If you would like to use ParaSCIP, you have to install an MPI library and have to set up MPI
   environment.
 
3. Create a folder lib/ in your ug directory and create a softlink there
   pointing to your SCIP directory.
 
4. Compile the UG project, by entering "make [options]" where the options are
   the same as in SCIP except COMM parameter. You can find SCIP and its documentation on
   http://scip.zib.de .
 
   To make FiberSCIP(default):
      "make [options] [COMM=pth]"
   To make ParaSCIP:
      "make [options] COMM=mpi"
 
5. You can run short testset as follows:
 
   To test FiberSCIP(default):
      "make [options] [COMM=pth] test"
   To test ParaSCIP:
      "make [options] COMM=mpi test"
 
USAGE:
 
Here you'll find the basic instructions how FiberSCIP and ParaSCIP can be used to solve MIPs
and MINLPs in parallel.
 
1. How to run FiberSCIP ("fscip" is a binary file name).
 
    "fscip fscip_param_file problem_file_name [-l <logfile>] [-q] [-sl <settings>] [-s <settings>] [-sr <root_settings>]
     [-w <prefix_warm>] [-sth <number>] [-fsol <solution_file>] [-isol <initial solution file]"
       -l <logfile>           : copy SCIP output into log file
       -q                     : suppress SCIP screen messages
       -sl <settings>         : load parameter settings (.set) file for LoadCoordinator presolving
       -sr <root_settings>    : load parameter settings (.set) file for root node subtree solving
       -s <settings>          : load parameter settings (.set) file for additional node solving
       -w <prefix_warm>       : warm start file prefix ( prefix_warm_nodes.gz and prefix_warm_solution.txt are read )
       -sth <number>          : the number of solver threads used
       -fsol <solution file> : specify output solution file
       -isol <intial solution file> : specify initial solution file
 
2. How to run ParaSCIP ("parascip" is a binary file name).
 
    "mpirun -np #MPI_processes(#solvers + 1) parascip parascip_param_file problem_file_name [-l <logfile>] [-q] [-sl <settings>] [-s <settings>] [-sr <root_settings>]
     [-w <prefix_warm>] [-sth <number>] [-fsol <solution_file>] [-isol <initial solution file]"
       -l <logfile>           : copy SCIP output into log file
       -q                     : suppress SCIP screen messages
       -sl <settings>         : load parameter settings (.set) file for LoadCoordinator presolving
       -sr <root_settings>    : load parameter settings (.set) file for root node subtree solving
       -s <settings>          : load parameter settings (.set) file for additional node solving
       -w <prefix_warm>       : warm start file prefix ( prefix_warm_nodes.gz and prefix_warm_solution.txt are read )
       -sth <number>          : the number of solver threads used
       -fsol <solution file> : specify output solution file
       -isol <intial solution file> : specify initial solution file
 
     [NOTE]
     If the instance you want to solve are not MIP, then you have to ensure that all MPI processes can access
     the instance data file. All ParaSolver processes accesses the file for the initialization.
 
ABOUT FiberSCIP and ParaSCIP PARAMETERS
 
The current default parameter settings are intended for using FiberSCIP on relatively small scale shared memory
computing environments with one or two hours timelimit. Using the default settings, the computing log is output
to standard out and solution files are saved to the working directory.  In order to save additional log information
the user should uncomment the line Quiet=FALSE in settings/default.set, statistical information will be saved to
log files in the working directory.  Alternatively, the log files can be saved to a user created directory 'logs'
by uncommenting additional lines in the settings file.
 
The UG framework specific parameters are mainly for controlling the load balancing and should be set by the user
in the main settings file.  Independent parameter settings files can be used to specify SCIP parameters to be used
at three different stages of the solution process (Load Coordinator presolving, root node subtree solving,
additional node solving).
The parameters have to be set taking into account:
- Number of parallel solvers (number of cores) used,
- Time and memory limitations,
- The instance to be solved.
 
While the default parameter settings are appropriate for 2-16 cores, tuning parameter settings for large-scale
environments is more challenging.  If you would like help tuning ParaSCIP on a large-scale environment to solve
hard instances, please e-mail shinano@zib.de with a log file from solving your instance for about two hours
along with information about your platform.
ParaSCIP has been successfully run on as many as 80,000 cores of the TITAN supercomputer and we are happy to
recommend parameter settings for other large-scale environments.
 
KNOWN-BUGS-AND-WORKAROUNDS
 
When you solve MINLPs, you may meet an abort. Then, please use the patch included in this archive.
It can be applied doing
   "patch -p0 < scip.patch"
from the root directory of ug.
 
In order to use a time limit, it is best to run FiberSCIP and ParaSCIP under a shell with a hard time limit.
Due to the limitations in the communication between the processes or threads it is difficult to terminate
cleanly when a soft time limit is used, possibly causing the process to hang.
 
LICENSING:
 
UG, ParaSCIP, FiberSCIP are licensed under the GNU Lesser General Public License; see the file COPYING.