Originally posted by: AMurshed

Hello,

I am trying to solve a cost function for Ethernet using CPLEX MIPS.

The program works and gives correct results for small data (10 nodes, 4 jobs and 4 messages)

The problem is that when I increase the number of jobs more it crashes and gives me 'Segmentation fault (core dumped)'.

Is there a way to fix the program or because CPLEX can not work for large data?

Thanks in advance,

Ayman Murshed

Hi,

can you attach a .sav you get a seg fault with?

regards

Originally posted by: AMurshed

Thanks Mr. Alex for your reply.

Attached is the cumulative.lp file for the small network with  the following input:

Jobs, messages, nodes[4,4,7]

network topology
[[0,1,1,1,0,0,1],
 [1,0,0,0,0,0,0],
 [1,0,0,0,0,0,0],
 [1,0,0,0,1,1,1],
 [0,0,0,1,0,0,0],
 [0,0,0,1,0,0,0],
 [1,0,0,1,0,0,0]]

sending jobs

 [0,0,1,2]     

recieving jobs
[[0, 1, 0, 0],
 [0, 0, 1, 0],
 [0, 0, 0, 1],
 [0, 1, 0, 0]]

Minimum arrival time
[50,50,50,50]

duration of hop-to-hop transmission
[3,3,3,3]

execution time for job
[2,2,2,2]

allocatability of nodes
[0,1,1,0,1,1,1]

The problem is that when the number of jobs or messages or nodes increased, it gives a segmentation fault.

Thanks in advence,

Best regards,

Originally posted by: EdKlotz

Ayman,

             I'm not clear regarding whether you attached an LP file that crashed, or the smaller model that ran without problem.    Regardless, I just ran your attached model with CPLEX 12.6.1 and default settings, and it finished the model with no problems.   Can you provide the CPLEX version you are using, and also confirm that the attached file is indeed the one on which you encountered the crash?   Also, when you encounter the crash, how are you using CPLEX?   Are you just using the interactive optimizer (i.e. the cplex executable), or are you running it from your own program?   If the latter, which API are you using?

                                                                                                                                                                                                                   Ed

Originally posted by: A_Murshed

Thanks EdKlotz for your reply.

Actually, the attached lp file is for small network and it's working fine with execution time =340 seconds.

The problem when I input large network (for example)

Jobs, messages, nodes[9,9,14]

network topology
[[0,1,1,1,1,1,1,0,0,0,1,0,0,0],
 [1,0,0,0,0,0,0,0,0,0,0,0,0,0],
 [1,0,0,0,0,0,0,0,0,0,0,0,0,0],
 [1,0,0,0,0,0,0,0,0,0,0,0,0,0],
 [1,0,0,0,0,0,0,0,0,0,0,0,0,0],
 [1,0,0,0,0,0,0,0,0,0,0,0,0,0],
 [1,0,0,0,0,0,0,1,1,1,1,0,0,0],
 [0,0,0,0,0,0,1,0,0,0,0,0,0,0],
 [0,0,0,0,0,0,1,0,0,0,0,0,0,0],
 [0,0,0,0,0,0,1,0,0,0,0,0,0,0],
 [1,0,0,0,0,0,1,0,0,0,0,1,1,1],
 [0,0,0,0,0,0,0,0,0,0,1,0,0,0],
 [0,0,0,0,0,0,0,0,0,0,1,0,0,0],
 [0,0,0,0,0,0,0,0,0,0,1,0,0,0]]

sending jobs

[1,2,3,4,5,6,7,8,8]

recieving jobs
[[1,0,0,0,0,0,0,0,0],
 [1,0,0,0,0,0,0,0,0],
 [0,1,0,0,0,0,0,0,0],
 [0,0,0,1,0,0,0,0,0],
 [1,0,0,0,0,0,0,0,0],
 [0,0,0,1,0,0,0,0,0],
 [0,0,0,0,0,0,1,0,0],
 [0,0,0,0,0,0,1,0,0],
 [0,0,0,0,0,0,0,1,0]]

Minimum arrival time
[80,80,80,80,80,80,80,80,80]

duration of hop-to-hop transmission
[2,2,2,2,2,2,2,2,2]

execution time for job
[1,1,1,1,1,1,1,1,1]

allocatability of nodes
[0,1,1,1,1,1,0,1,1,1,0,1,1,1]

It crashes before solving the model and doesn't create lp file.

I am using CPLEX enterprise Server 12.6 on my university (Siegen University) Unix Server.

Best regards,

Ayman Murshed

 

Hi,

if you use CPLEX Enterprise Server, can you attach the .mod and .dat files you use to get this SIGSERV?

regards

Originally posted by: A_Murshed

Hello,

I am sorry, I don't know what are those files (.mod and .dat).

I used c++ code format to write CPLEX program.

Is it necessary to attach the C++ code. The work in this code is not published yet and I am really afraid to post the code here.

Is there another way to solve the problem?

Regards,

Originally posted by: T_O

Are you sure that the segmentation fault comes from CPLEX and not from your C++ code? Did you try to compile with debugging symbols and run the whole thing inside a debugger?

Best regards,
Thomas

Originally posted by: A_Murshed

Thanks Thomas for your reply,

Yes, the program is working very well and produce correct results for small network topology (attached an lp file in my second reply).

The problem is that when I increase the input arrays of the network topology.

Regards,

Originally posted by: T_O

Unfortunately, this does not mean anything. If your memory allocation routine is erroneous, anything might happen. So please run your program through a debugger and check where the segmentation fault comes from. You might also try to set the datacheck-parameter to 1.

Best regards,
Thomas

Originally posted by: A_Murshed

This is part of the code for Numerical arrays and variables arrays:

///////////////////////////////////////////////////////////////

#include "ilcplex/ilocplex.h"
ILOSTLBEGIN

typedef IloArray<IloBoolVarArray>  BoolVarray2dim;
typedef IloArray<IloNumVarArray>   NumVarray2dim;
typedef IloArray<IloNumArray>   NumArray2dim;


int
main(int argc, char **argv)
{
   IloEnv env;
   try {

     ///////////Constants//////////

const char* filename;     

     IloNumArray JMR(env);  //Network Dimension (Jobs, Messages, Routers)
     NumArray2dim CONN(env);               //CONN = Connection    
     IloNumArray SENDER(env);  // Source Job of Messages         
     NumArray2dim DEST(env);               //DEST = Destination         
     IloNumArray MINT(env);  // Minimum Interarrival time of Sporadic Msg., period of periodic msg.
     IloNumArray DUR(env);    // Duration of message transmission (depending on message lenght).
     IloNumArray EXEC(env);   // Computational time of job
     IloBoolArray ALLOCABILITY(env);  //Allocability array

     filename = "Data/10N.dat";
      ifstream f(filename, ios::in);
      if (!f) {
         cerr << "No such file: " << filename << endl;
         throw(1);
      }

      f >> JMR;
      f >> CONN;
      f >> SENDER;
      f >> DEST;      
      f >> MINT;
      f >> DUR;
      f >> EXEC;
      f >> ALLOCABILITY;



     ////////Decision Variables///////
     IloNumVarArray ALLOC(env,JMR[0], 0,JMR[2],ILOINT);  // Allocation of jobs to cores (one core connected to one router)     
     IloNumVarArray   HOPS(env, JMR[1], 0, JMR[2], ILOINT);               // Number of hobs for each message
     
     NumVarray2dim PATH(env,JMR[1]);         // Each row is the path for a message.
     for(IloInt i =0; i<JMR[1];++i)
         PATH[i]= IloNumVarArray(env,JMR[2], 0, JMR[2], ILOINT);
     
    BoolVarray2dim ALLOCMATRIX(env,JMR[0]);         // Allocation of jobs to cores.
     for(IloInt i =0; i<JMR[2];++i)
         ALLOCMATRIX[i]= IloBoolVarArray(env,JMR[2]);

    IloNumVarArray INJECTIME(env,JMR[1], 0,MINT[0],ILOINT);
   
     
     NumVarray2dim ONPATH(env,JMR[1]); //ONPATH = Message meets router     
     for(IloInt i=0;i<JMR[1];++i)
         ONPATH[i]= IloNumVarArray(env,JMR[2], 0,1, ILOINT);
 
     IloModel modl(env);
     IloCplex cplex(env);

//Objective ////MINIMIZE cost////////
 IloExpr objExp(env,0);
      for(IloInt m=0;m<JMR[1];++m)
         objExp += INJECTIME[m] + HOPS[m]*DUR[m];      
      modl.add(IloMinimize(env,objExp));
      objExp.end();

Originally posted by: T_O

Please use a debugger (e.g. gdb) to see where the segmentation fault comes from. It is really easy.

g++ -g [...]
gdb [executable]
r [optinal parameters of the executable]
(wait for segmentation fault)
bt

Best regards,
Thomas

Originally posted by: A_Murshed

This is the output of the suggested debug:

(gdb) r
Starting program: /home/murshed/cplex_test/test
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".

Program received signal SIGSEGV, Segmentation fault.
0x000000000040931e in main ()

(gdb) bt
#0  0x000000000040931e in main ()

Originally posted by: T_O

Please recompile with debugging symbols (-g) and without optimization. Then try again.

Best regars,
Thomas

Originally posted by: A_Murshed

Hello ,

I used the command:

make --debug [executable]

murshed@opnet:~/cplex_test$ gdb test
GNU gdb (Ubuntu 7.7.1-0ubuntu5~14.04.2) 7.7.1
Copyright (C) 2014 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
and "show warranty" for details.
This GDB was configured as "x86_64-linux-gnu".
Type "show configuration" for configuration details.
For bug reporting instructions, please see:
<http://www.gnu.org/software/gdb/bugs/>.
Find the GDB manual and other documentation resources online at:
<http://www.gnu.org/software/gdb/documentation/>.
For help, type "help".
Type "apropos word" to search for commands related to "word"...
Reading symbols from test...(no debugging symbols found)...done.
(gdb) r
Starting program: /home/murshed/cplex_test/test
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".

Program received signal SIGSEGV, Segmentation fault.
0x000000000040931e in main ()
(gdb) bt
#0  0x000000000040931e in main ()
 

Originally posted by: T_O

As far as I know, --debug does not have anything to do with debugging symbols. Can you try something like

CFLAGS="-g -O0" CXXFLAGS="-g -O0" make

Or even better: Post your makefile.

Best regards,
Thomas

Originally posted by: A_Murshed

Attached, makefile.

Originally posted by: A_Murshed

I tried also the parameters you suggested, but gave the same results.

Originally posted by: T_O

Thanks. In your makefile, replace

CCOPT =  -O -fPIC -fexceptions -DNDEBUG -DIL_STD

by

CCOPT =  -g -fPIC -fexceptions -DIL_STD

and try again.

Best regards,
Thomas

Originally posted by: A_Murshed

Thanks Mr. Thomas for your reply,

That's the output after modifying the makefile:

murshed@opnet:~/cplex_test$ gdb test
GNU gdb (Ubuntu 7.7.1-0ubuntu5~14.04.2) 7.7.1
Copyright (C) 2014 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
and "show warranty" for details.
This GDB was configured as "x86_64-linux-gnu".
Type "show configuration" for configuration details.
For bug reporting instructions, please see:
<http://www.gnu.org/software/gdb/bugs/>.
Find the GDB manual and other documentation resources online at:
<http://www.gnu.org/software/gdb/documentation/>.
For help, type "help".
Type "apropos word" to search for commands related to "word"...
Reading symbols from test...done.

(gdb) r
Starting program: /home/murshed/cplex_test/test
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
X& IloArray::operator[] (IloInt i) : Out of bounds operation: index superior to size of array
test: include/ilconcert/iloenv.h:2219: X& IloArray<X>::operator[](IloInt) [with X = IloBoolVarArray; IloInt = long int]: Assertion `(i < _impl->getSize()) || (std:: cerr << "X& IloArray::operator[] (IloInt i) : Out of bounds operation: index superior to size of array" << std:: endl, ilo_stop_assert())' failed.

Program received signal SIGABRT, Aborted.
0x00007ffff700cbb9 in __GI_raise (sig=sig@entry=6)
    at ../nptl/sysdeps/unix/sysv/linux/raise.c:56
56      ../nptl/sysdeps/unix/sysv/linux/raise.c: No such file or directory.
(gdb) bt
#0  0x00007ffff700cbb9 in __GI_raise (sig=sig@entry=6)
    at ../nptl/sysdeps/unix/sysv/linux/raise.c:56
#1  0x00007ffff700ffc8 in __GI_abort () at abort.c:89
#2  0x00007ffff7005a76 in __assert_fail_base (
    fmt=0x7ffff71572b0 "%s%s%s:%u: %s%sAssertion `%s' failed.\n%n",
    assertion=assertion@entry=0xfbf228 "(i < _impl->getSize()) || (std:: cerr << \"X& IloArray::operator[] (IloInt i) : Out of bounds operation: index superior to size of array\" << std:: endl, ilo_stop_assert())",
    file=file@entry=0xfbe589 "include/ilconcert/iloenv.h",
    line=line@entry=2219,
    function=function@entry=0xfbf920 <IloArray<IloBoolVarArray>::operator[](long)::__PRETTY_FUNCTION__> "X& IloArray<X>::operator[](IloInt) [with X = IloBoolVarArray; IloInt = long int]") at assert.c:92
#3  0x00007ffff7005b22 in __GI___assert_fail (
    assertion=0xfbf228 "(i < _impl->getSize()) || (std:: cerr << \"X& IloArray::operator[] (IloInt i) : Out of bounds operation: index superior to size of array\" << std:: endl, ilo_stop_assert())",
    file=0xfbe589 "include/ilconcert/iloenv.h", line=2219,
    function=0xfbf920 <IloArray<IloBoolVarArray>::operator[](long)::__PRETTY_FUNCTION__> "X& IloArray<X>::operator[](IloInt) [with X = IloBoolVarArray; IloInt = long int]") at assert.c:101
#4  0x000000000040f6fe in IloArray<IloBoolVarArray>::operator[] (
    this=0x7fffffffe020, i=8) at include/ilconcert/iloenv.h:2219
 

Regards,

Originally posted by: A_Murshed

By the way, it gives me this output even if I put small input values not only for large input values.

Originally posted by: T_O

This is what I guessed. Some array dimensions mismatch and in some runs, you just seem to be lucky not to run into trouble.

Is there some more output? Or did you write down all lines? What happens if you type

where

instead of

bt

? We still have not found the root of your problem. Could you also check the dimensions in your input files?

Best regards,
Thomas

Originally posted by: A_Murshed

Thanks Mr. Thomas,

I think all arrays are well defined:

     IloNumArray JMR(env);  //Network Dimension (Jobs, Messages, Routers)
     NumArray2dim CONN(env);               //CONN = Connection    
     IloNumArray SENDER(env);  // Source Job of Messages         
     NumArray2dim DEST(env);               //DEST = Destination         
     IloNumArray MINT(env);  // Minimum Interarrival time of Sporadic Msg., period of periodic msg.
     IloNumArray DUR(env);    // Duration of message transmission (depending on message lenght).
     IloNumArray EXEC(env);   // Computational time of job
     IloBoolArray ALLOCABILITY(env);  //Allocability array

     filename = "Data/10N.dat";
      ifstream f(filename, ios::in);
      if (!f) {
         cerr << "No such file: " << filename << endl;
         throw(1);
      }

      f >> JMR;
      f >> CONN;
      f >> SENDER;
      f >> DEST;      
      f >> MINT;
      f >> DUR;
      f >> EXEC;
      f >> ALLOCABILITY;

and this is the variables:

 IloNumVarArray ALLOC(env,JMR[0], 0,JMR[2],ILOINT);  // Allocation of jobs to cores (one core connected to one router)     
     IloNumVarArray   HOPS(env, JMR[1], 0, JMR[2], ILOINT);               // Number of hobs for each message
     
     NumVarray2dim PATH(env,JMR[1]);         // Each row is the path for a message.
     for(IloInt i =0; i<JMR[1];++i)
         PATH[i]= IloNumVarArray(env,JMR[2], 0, JMR[2], ILOINT);
     
    BoolVarray2dim ALLOCMATRIX(env,JMR[0]);         // Allocation of jobs to cores.
     for(IloInt i =0; i<JMR[2];++i)
         ALLOCMATRIX[i]= IloBoolVarArray(env,JMR[2]);

    IloNumVarArray INJECTIME(env,JMR[1], 0,MINT[0],ILOFLOAT);
    IloNumVarArray OBJECTI(env,JMR[1], 0.0,1000.0,ILOFLOAT);
     
     NumVarray2dim ONPATH(env,JMR[1]); //ONPATH = Message meets router     
     for(IloInt i=0;i<JMR[1];++i)
         ONPATH[i]= IloNumVarArray(env,JMR[2], 0,1, ILOINT);
 

Attached the input file.

Best regards,

Originally posted by: A_Murshed

This is the output after "where" command:

(gdb) r
Starting program: /home/murshed/cplex_test/test
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
X& IloArray::operator[] (IloInt i) : Out of bounds operation: index superior to size of array
test: include/ilconcert/iloenv.h:2219: X& IloArray<X>::operator[](IloInt) [with X = IloNumVarArray; IloInt = long int]: Assertion `(i < _impl->getSize()) || (std:: cerr << "X& IloArray::operator[] (IloInt i) : Out of bounds operation: index superior to size of array" << std:: endl, ilo_stop_assert())' failed.

Program received signal SIGABRT, Aborted.
0x00007ffff700cbb9 in __GI_raise (sig=sig@entry=6)
    at ../nptl/sysdeps/unix/sysv/linux/raise.c:56
56      ../nptl/sysdeps/unix/sysv/linux/raise.c: No such file or directory.
(gdb) where
#0  0x00007ffff700cbb9 in __GI_raise (sig=sig@entry=6)
    at ../nptl/sysdeps/unix/sysv/linux/raise.c:56
#1  0x00007ffff700ffc8 in __GI_abort () at abort.c:89
#2  0x00007ffff7005a76 in __assert_fail_base (
    fmt=0x7ffff71572b0 "%s%s%s:%u: %s%sAssertion `%s' failed.\n%n",
    assertion=assertion@entry=0xfbea98 "(i < _impl->getSize()) || (std:: cerr << \"X& IloArray::operator[] (IloInt i) : Out of bounds operation: index superior to size of array\" << std:: endl, ilo_stop_assert())",
    file=file@entry=0xfbde89 "include/ilconcert/iloenv.h",
    line=line@entry=2219,
    function=function@entry=0xfbf1a0 <IloArray<IloNumVarArray>::operator[](long)::__PRETTY_FUNCTION__> "X& IloArray<X>::operator[](IloInt) [with X = IloNumVarArray; IloInt = long int]") at assert.c:92
#3  0x00007ffff7005b22 in __GI___assert_fail (
    assertion=0xfbea98 "(i < _impl->getSize()) || (std:: cerr << \"X& IloArray::operator[] (IloInt i) : Out of bounds operation: index superior to size of array\" << std:: endl, ilo_stop_assert())",
    file=0xfbde89 "include/ilconcert/iloenv.h", line=2219,
    function=0xfbf1a0 <IloArray<IloNumVarArray>::operator[](long)::__PRETTY_FUNCTION__> "X& IloArray<X>::operator[](IloInt) [with X = IloNumVarArray; IloInt = long int]") at assert.c:101
#4  0x000000000040f016 in IloArray<IloNumVarArray>::operator[] (
    this=0x7fffffffe020, i=4) at include/ilconcert/iloenv.h:2219


 

Originally posted by: T_O

I have to admit that I am not sure whether your arrays contain what you intend them to contain after these commands (I typically use the C API, not concert):

      f >> JMR;
      f >> CONN;
      f >> SENDER;
      f >> DEST;      
      f >> MINT;
      f >> DUR;
      f >> EXEC;
      f >> ALLOCABILITY;

So, could you please look into these arrays and check whether they look as intended and have the correct size?

Originally posted by: A_Murshed

Yes, I printed out the input arrays and they are the same as they look in the input file.

Originally posted by: T_O

Hmm, I am still wondering why the backtrace does not end in the main method. Could you try

thread apply all backtrace

Originally posted by: A_Murshed

Dear Mr. Thomas,

I finally managed to find the line of code that did the segmentation fault.

In my sixth post, I wrote:

BoolVarray2dim ALLOCMATRIX(env,JMR[0]);         // Allocation of jobs to cores.
     for(IloInt i =0; i<JMR[2];++i)
         ALLOCMATRIX[i]= IloBoolVarArray(env,JMR[2]);

I defined an array with JMR[0] values and then use JMR[2] values.

Now the program works but takes a lot of time. At least the SEGSEGV error disappeared.

Many thanks for your help and my silly mistake.

Regards,

Ayman

Originally posted by: T_O

Remember to use the original makefile. This might speed it up a little again.

Best regards,
Thomas

Originally posted by: EdKlotz

Ayman,

              Since you appear to be on Linux, try running the program, compiled with debug options as Thomas has explained, with Valgrind, a memory checker.    This will have the advantage if identifying the problem right when it happens, whereas running with gdb tends to provide info at the crash point.    Sometimes the crash occurs very close to the problematic line of code, but in many other cases it occurs much later.   For the latter cases, valgrind can be much more effective than gdb.   Valgrind is free and often comes with Linux distributions; try just typing the 'valgrind' command and see if you get something on your machine.   Otherwise, go to http://www.valgrind.org/ for downloads as well as additional docs beyond what is available with the --help command line option.

               Another useful Linux debugging tool is the Undo debugger.   It's not free, but it can save a lot of time, as it allows you to move both forward and backward in your gdb session.    The URL is http://undo-software.com/.

Overall, I agree with Thomas that the most likely cause of the trouble here is some inconsistency between the dimensioning of your arrays in your C++ program and either input data or subsequent lines of code that go past the array bounds.   However, if you export LP and SAV files of the problematic model and can reproduce crashes in both files in interactive CPLEX, then you should post those files to this thread.

Originally posted by: AMurshed

Thanks Mr. Thomas and Mr. Ed.

The program now keeps on searching for a solution and takes tooo much time to process (more than 10000 s) and then gives me  (killed) output !?

I have upgraded the CPLEX into 12.6.1 installed on multiprocessor (4 processors) server of type (Intel(R) Xeon(R) CPU E5-2430 0 @ 2.20GHz)

Can anyone gives me a hint on how to speed-up the solution time

If you get the message 'killed' for a program on a Linux/Unix machine then this usually means that your program ran out of memory and was killed by the operating system for that reason.

To learn how to fight memory issues please read this chapter in the user manual.

When trying to speed up the solution times, have you tried the tuning tool?

Originally posted by: AMurshed

Thanks Mr. Daniel,

Could you please suggest any optimization code.

I have tried to use :

       cplex.setParam(IloCplex::VarSel, 3);
       cplex.setParam(IloCplex::Probe, 3);
       cplex.setParam(IloCplex::NodeSel, 2) ;
       cplex.setParam(IloCplex::BBInterval, 0) ;
       cplex.setParam(IloCplex::BtTol, 0.0) ;

But still it keeps working without finding solution.

Originally posted by: EdKlotz

Ayman,

>>>>>>>>

I have tried to use :

       cplex.setParam(IloCplex::VarSel, 3);
       cplex.setParam(IloCplex::Probe, 3);
       cplex.setParam(IloCplex::NodeSel, 2) ;
       cplex.setParam(IloCplex::BBInterval, 0) ;
       cplex.setParam(IloCplex::BtTol, 0.0) ;

>>>>>>>>

What prompted you to try these non-default parameters?   Did you try CPLEX with just default parameter settings, and the results were unsatisfactory?    Specifically, if CPLEX has trouble finding any feasible solutions at all, leave the backtrack tolerance (IloCplex::BtTol) at its default rather than setting it to 0 as you do above.   I would also suggest trying a run with the variable selection left at its default rather than set to the computationally expensive setting of 3 (strong branching).

As Daniel pointed out, the manual has a chapter on dealing with out of memory issues.   Try setting the file parameter to 3 to help in this regard.

If you still have trouble, please include a CPLEX node log of the slow run in your next post.

Originally posted by: AMurshed

Thanks Mr. Ed for the explainantion,

Now, I tried to change the GAP to 0.01:    cplex.setParam(IloCplex::EpGap, 0.01);

The best bound is changing slowly.

Attached an lp file.

Thanks in advence.

Originally posted by: EdKlotz

Ayman,

              I ran CPLEX 12.6.0.1 on my laptop with default settings.    While performance wasn't sspectacular for a model of such modest size, CPLEX did solve it to optimality in under 14 minutes:

  16814   351    infeasible             16.0000       10.0000  1850440   37.50%
  17067   116    infeasible             16.0000       13.0000  1945950   18.75%
  17145    40       13.0000  4018       16.0000       13.0000  1980510   18.75%
Elapsed time = 750.29 sec. (288770.36 ticks, tree = 0.01 MB, solutions = 4)
  17146    41       13.0000  5498       16.0000       13.0000  1986963   18.75%
  17180    12       13.0000  4104       16.0000       13.0000  2011550   18.75%
  17186    12       13.0000  5140       16.0000       13.0000  2026351   18.75%
  17195     5    infeasible             16.0000       13.0000  2041997   18.75%
  17202     4       14.0000  5812       16.0000       14.0000  2056216   12.50%

GUB cover cuts applied:  285
Clique cuts applied:  5307
Cover cuts applied:  2963
Implied bound cuts applied:  690
Flow cuts applied:  273
Mixed integer rounding cuts applied:  673
Zero-half cuts applied:  675
Gomory fractional cuts applied:  48

Root node processing (before b&c):
  Real time             =   24.80 sec. (9493.49 ticks)
Parallel b&c, 8 threads:
  Real time             =  795.81 sec. (317984.43 ticks)
  Sync time (average)   =  210.61 sec.
  Wait time (average)   =    0.00 sec.
                          ------------
Total (root+branch&cut) =  820.61 sec. (327477.93 ticks)

Solution pool: 4 solutions saved.

MIP - Integer optimal solution:  Objective = 1.6000000000e+001
Solution time =  820.61 sec.  Iterations = 2060514  Nodes = 17207
Deterministic time = 327477.97 ticks  (399.07 ticks/sec)

CPLEX>

What version of CPLEX are you using, and how does this compare to when you run with default settings?

Originally posted by: AMurshed

Thanks Ed for your reply,

I use IBM ILOG CPLEX Optimization Studio V12.6.1 installed on multiprocessor (12 processors) server of type (Intel(R) Xeon(R) CPU E5-2430 0 @ 2.20GHz).

The problem is that the (Best Bound) changed very slowly  and some times it stops without any reason for large problem.

Can you please try the attached lp file and see if it executes and finishes fast , I return all parameters to its default values.

The problem consists of small network: (4 Jobs, 4 Messages, and 13 nodes)

Thanks in advance,

Originally posted by: AMurshed

Any suggestions?

Originally posted by: AMurshed

Hello,

I am sorry for asking too many questions.

I just want to know if the order to the constraints plays a significant rule the gives different execution time ???

Thanks in advance.

Ayman

Yes, the order of constraints (or variables) can play a significant role in the execution time. This has been discussed many times on this Forum and you may find more details by searching the Forum.

Originally posted by: AMurshed

Thanks Daniel for your answer.

I've finally managed to find the answer on how to decrease the execution time of my CPLEX model.

I think CPLEX tries to look for all possible paths along while finding solution(s).

The main drawback of CPLEX is one need to narrow the range of possible variables values in order to let CPLEX find the answer in this defined area. Otherwise, CPLEX will look at all (possible+impossible) paths and then gives you the solution after couple of thousands of seconds.

I think when compared to SAT solver, this is the main drawback of CPLEX solver.

Originally posted by: AMurshed

Is there a possibility that one can play with the Heuristic search in CPLEX?

I would like to change the way CPLEX uses the Heuristic search so that I could get better execution time.

In general the frequency with which heuristics are applied is defined by CPX_PARAM_HEURFREQ.

There are also parameters to explicitly control some heuristics, you can find them by looking through the list of CPLEX parameters.