Compile Pagmo2.19 for MPI support?

[montanaviking]

Hi,
I just built Pagmo 2.19 with the following make options (done from /xxx/build directory as recommended in the build instructions:
$ sudo cmake ../ -DPAGMO_BUILD_TESTS=ON -DPAGMO_BUILD_TUTORIALS=ON -DPAGMO_WITH_EIGEN3=ON -DPAGMO_WITH_NLOPT=ON -DPAGMO_WITH_IPOPT=ON -DENABLE_MPI=ON -DCMAKE_INSTALL_PREFIX=/opt/pagmo

This builds and apparently runs with no issues, except as discussed NLOPT was not found (previous discussion) but the NLOPT problem was solved as per an earlier discussion from me.

Now, I'm wondering how to enable MPI support? I followed the instructions on https://esa.github.io/pagmo/mpi_support.html
enabling the MPI support for cmake as shown above. However, it appears that there is no mpi_island.hpp generated in my built includes directory: /opt/pagmo/include/pagmo/islands

My questions are:

1. Is there MPI support in Pagmo 2.19?
2. If so, could someone kindly tell me how to enable it and/or how to check if it's enabled?
   Thanks in advance!
   Phil

[montanaviking]

Hi,
The answer appears to be in:
https://assets.researchsquare.com/files/rs-2433869/v1_covered.pdf
and according to these authors, MPI cluster computing was available in Pagmo 1x but not in Pagmo 2.x.
However, Pagmo 2.x does support multithreading within a single machine via Intel tbb library and I do use that.
Please correct me if I'm wrong.
I'm going to be attempting to implement cluster computation by the use of TCP sockets (using the Qt6.x libraries) and Boost or another serialization method.
If anyone's interested, please let me know.
Phil

[montanaviking]

Also, I'd like to work towards adding MPI or equivalent capability for running Pagmo2 on clusters. I'm wondering how to approach this? I'd like to have islands perform migrations between islands running on different machines. Where to start?
TBB works very well handling thread pools in Pagmo and I make extensive use of this. I'm looking at using multiple machines to expand the island populations using the grey wolf and/or multi basin hopping optimizations using TCP sockets and something like Boost serialization to communicate between machines running independent populations on each machine, but I suspect it would work better if I could actually perform island migrations without confining the migration to the machine the island is running on - i.e. accommodating island migrations between machines as well.
Anyone else interested in this?
Thanks and best,
Phil