Originally posted by: C.Schmidt

Dear forum,

I have developed a model for a customer and now need to give the hardware recommendation for the solver machine. The solver ran on win 7 64 Bit with a core i7 (quadcore with hyperthreading - 8 virtual cores) and 16 GB RAM. Please find below the log of an example run. The model was solved within 26 hours. During the run I observed that there were long periods with only 1-2 cores used (so Intel's Turbo boost should have helped) and some time with all 8 cores used. Reserved and private memory was around 16 GB.

Will the solver significantly benefit from a dual core i7 system? Will it help adding more RAM?
What is your experience/recommendation?

BTW I'll post questions on how to improve the solution speed by modifying the model or solver settings in a different thread.

Thanks for any advice,

Christian
Log:

Welcome to IBM(R) ILOG(R) CPLEX(R) Interactive Optimizer 12.2.0.0
  with Simplex, Mixed Integer & Barrier Optimizers
5725-A06 5725-A29 5724-Y48 5724-Y49 5724-Y54 5724-Y55
Copyright IBM Corp. 1988, 2010.  All Rights Reserved.
 
Type 'help' for a list of available commands.
Type 'help' followed by a command name for more
information on commands.
 
CPLEX> Problem 'C:\Users\appserv\AppData\Local\Temp\Schedule.lp' read.
Read time =    2.11 sec.
CPLEX> Tried aggregator 2 times.
MIP Presolve eliminated 33670 rows and 401085 columns.
MIP Presolve modified 3219 coefficients.
Aggregator did 1083 substitutions.
Reduced MIP has 26910 rows, 251615 columns, and 642617 nonzeros.
Reduced MIP has 16175 binaries, 0 generals, 0 SOSs, and 0 indicators.
Presolve time =    2.92 sec.
Found feasible solution after 3.15 sec.  Objective = 1793363.6000
Probing time =    0.31 sec.
Clique table members: 415.
MIP emphasis: balance optimality and feasibility.
MIP search method: dynamic search.
Parallel mode: deterministic, using up to 8 threads.
Root relaxation solution time =   81.53 sec.
 
        Nodes                                         Cuts/ 
   Node  Left     Objective  IInf  Best Integer     Best Node    ItCnt     Gap
 
*     0+    0                      1793363.6000                 158186     ---
*     0+    0                      1793000.0000                 158186     ---
      0     0   164791.1199  1495  1793000.0000   164791.1199   158186   90.81%
      0     0   165465.3241  1756  1793000.0000    Cuts: 2502   421742   90.77%
CPLEX Error  1256: Basis singular.
*     0+    0                       262234.8044   165465.3241  1346430   36.90%
      0     0   165598.4213  1614   262234.8044    Cuts: 2970  2328568   36.85%
*     0+    0                       226644.1113   165598.4213  3363961   26.93%
      0     0   165809.6578  1732   226644.1113    Cuts: 2906  3363961   26.84%
CPLEX Error  1256: Basis singular.
CPLEX Error  1256: Basis singular.
*     0+    0                       221103.5302   165809.6578  3363961   25.01%
CPLEX Error  1256: Basis singular.
      0     0        cutoff         221103.5302                4472392    0.00%
Elapsed real time = 95569.38 sec. (tree size =  0.00 MB, solutions = 5)
 
GUB cover cuts applied:  134
Clique cuts applied:  2
Cover cuts applied:  5
Implied bound cuts applied:  875
Flow cuts applied:  2128
Zero-half cuts applied:  10
Gomory fractional cuts applied:  20
 
Root node processing (before b&c):
  Real time             = 95566.46
Parallel b&c, 8 threads:
  Real time             =    0.00
  Sync time (average)   =    0.00
  Wait time (average)   =    0.00
                          -------
Total (root+branch&cut) = 95566.46 sec.
 
Solution pool: 5 solutions saved.
 
MIP - Integer optimal solution:  Objective = 2.2110353018e+005
Solution time = 95569.44 sec.  Iterations = 4472392  Nodes = 0

#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

Wow, this is interesting. It takes more than one day to solve the root node although it does not seem too big. Would you be willing to share the model with us CPLEX developers so that we can check whether we can improve our code and save some time? If so, can you please export the model as .sav file and send it do daniel(dot)junglas(at)de(dot)ibm(dot)com.

The optimal hardware configuration is hard to tell before we know what exactly is consuming so much time. What version of CPLEX are you using? Version 12.2 or an older version?
I think hardware tuning should be kind of last resort. First you should check if you can improve your formulation or change CPLEX parameters to solve the model faster.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: C.Schmidt

Sorry for the delay. I've sent the model to you.

My question about the optimal hardware configuration is twofold: One is I'd like to estimate if and if how much the system would benefit from more cores or RAM. Second is I need to recommend hardware to my customer...

Independently of the hardware question I would certainly like to improve the formulation and do some tuning. What really surprised me was that the model has been solved in the root node although I have a lot of binaries and SOS1 constraints. A model were I removed the binaries and had only SOS1 hasn't been solved within 1 week.

BTW I used CPLEX 12.2.
#CPLEXOptimizers
#DecisionOptimization