mpi branch not working

[kibaekkim]

With mpi branch, we cannot run with more than a process. If we run with a process, the dual decomposition does not converge.

[michel2323]

Which example is that with how many processes?

You need:

• structjump branch of BundleMethod
• duals branch of StructJuMPSolverInterface
• and run without a remainer in the partitioning. (can be fixed)
• the mpi branch does only support ProximalDualMethod I can work on this in the GO Challenge code.

[kibaekkim]

I have tested dcap.jl with use_admm=false in line 104.

diff --git a/examples/dcap.jl b/examples/dcap.jl
index 0ab3ed9..9db0535 100644
--- a/examples/dcap.jl
+++ b/examples/dcap.jl
@@ -101,4 +101,4 @@ end
 # return the array of nonanticipativity variables
 nonanticipativity_vars() = [:x,:u]

-main_dcap(2,3,3,20; use_admm=true)
+main_dcap(2,3,3,20; use_admm=false)

Here is my environment:

• julia1.1.0
• BundleMethod#structjump
• JuDD#mpi
• StructJuMP#julia0.7
• StructJuMPSolverInterface#duals

I do not see any bundle is added to the master problem.

[michel2323]

It runs fine here in 19 iterations with the latest commit.

[kibaekkim]

works now