Decision Optimization

 View Only

is this the correct configuration for distributed parallel optimization

  • 1.  is this the correct configuration for distributed parallel optimization

    Posted Sun January 17, 2021 11:54 AM
    Hi
    I got 1 master and 5 slaves.
    I want a fully distributed solution so that the MIP will be solved 5 times faster  ( ideally speaking).
    I know the speedup will be less than 5 usually.

    On the master  I run :
    > read cfg.vmc
    > set DistMIP  Rampup Duration  1
    > read  protfold.mps
    > mipopt

    The cfg.vmc file looks like this , where I am showing only 2 out of the 5  slaves  (the other 3 slaves are similar. The IP addresses and machine names are fake).

    <vmc>

    <machine name="n01">
    <transport type="TCP/IP">
    <address host="200.300.400.101" port="12345"/>
    </transport>
    </machine>

    <machine name="n02">
    <transport type="TCP/IP">
    <address host="200.300.400.102" port="12345"/>
    </transport>
    </machine>
     
    </vmc>

    Each slave is started before starting the master , using command lines  :

    /home/myname/Cplex_1210_Shared/cplex/bin/x86-64_linux/cpxworker -protocol=tcpip -libpath=/home/myname/Cplex_1210_Shared/cplex/bin/x86-64_linux/ -address=200.300.400.101:12345

    /home/myname/Cplex_1210_Shared/cplex/bin/x86-64_linux/cpxworker -protocol=tcpip -libpath=/home/myname/Cplex_1210_Shared/cplex/bin/x86-64_linux/ -address=200.300.400.102:12345

    So I think my procedure is correct. My only concern is that I see no progress log on any of the slaves.

    Please confirm that my procedure is correct, and how to get the node log on each slave.


    I see this on the master :

    Presolve time = 0.03 sec. (12.46 ticks)
    Running distributed MIP on 5 solvers.
    Setting up 5 distributed solvers.
    Setup time = 0.09 sec. (0.01 ticks)
    Starting ramp-up.

    Version identifier: 12.10.0.0 | 2019-11-26 | 843d4de
    CPXPARAM_Simplex_Display 0
    1043 0
    CPXPARAM_Preprocessing_Linear 0
    CPXPARAM_Sifting_Display 0
    CPXPARAM_RandomSeed 201919284
    1165 0
    1166 0
    2145 0
    2159 1
    CPXPARAM_DistMIP_Rampup_Duration 1
    CPXPARAM_Barrier_Display 0
    CPXPARAM_Network_Display 0
    Probing time = 0.02 sec. (4.42 ticks)
    Detecting symmetries...
    Probing time = 0.01 sec. (4.29 ticks)
    Clique table members: 11142.
    MIP emphasis: balance optimality and feasibility.
    MIP search method: dynamic search.
    Parallel mode: deterministic, using up to 32 threads.
    Root relaxation solution time = 0.23 sec. (158.77 ticks)


    Root node processing (before b&c):
    Real time = 3.18 sec. (2792.49 ticks)
    Parallel b&c, 32 threads:
    Real time = 55.79 sec. (37541.53 ticks)
    Sync time (average) = 13.27 sec.
    Wait time (average) = 0.08 sec.
    ------------
    Total (root+branch&cut) = 58.97 sec. (40334.02 ticks)
    Ramp-up : worker 0 terminated with lpstat 113
    Ramp-up : worker 3 terminated with lpstat 113

    Ramp-up finished (winner: 3).
    Ramp-up time = 59.03 sec. (40334.10 ticks)
    Setting up 5 solvers for distributed tree search.
    Setup time = 0.11 sec. (0.01 ticks)
    Starting distributed tree search on 5 solvers.





    ------------------------------
    srinivas tamvada
    ------------------------------

    #DecisionOptimization