8.2.7. Using LAMMPS in client/server mode
Client/server coupling of two codes is where one code is the “client” and sends request messages to a “server” code. The server responds to each request with a reply message. This enables the two codes to work in tandem to perform a simulation. LAMMPS can act as either a client or server code.
Some advantages of client/server coupling are that the two codes run as stand-alone executables; they are not linked together. Thus neither code needs to have a library interface. This often makes it easier to run the two codes on different numbers of processors. If a message protocol (format and content) is defined for a particular kind of simulation, then in principle any code that implements the client-side protocol can be used in tandem with any code that implements the server-side protocol, without the two codes needing to know anything more specific about each other.
A simple example of client/server coupling is where LAMMPS is the client code performing MD timestepping. Each timestep it sends a message to a server quantum code containing current coords of all the atoms. The quantum code computes energy and forces based on the coords. It returns them as a message to LAMMPS, which completes the timestep.
Alternate methods for code coupling with LAMMPS are described on the Howto couple doc page.
LAMMPS support for client/server coupling is in its MESSAGE package which implements several commands that enable LAMMPS to act as a client or server, as discussed below. The MESSAGE package also wraps a client/server library called CSlib which enables two codes to exchange messages in different ways, either via files, sockets, or MPI. The CSlib is provided with LAMMPS in the lib/message dir. The CSlib has its own website with documentation and test programs.
For client/server coupling to work between LAMMPS and another code, the other code also has to use the CSlib. This can sometimes be done without any modifications to the other code by simply wrapping it with a Python script that exchanges CSlib messages with LAMMPS and prepares input for or processes output from the other code. The other code also has to implement a matching protocol for the format and content of messages that LAMMPS exchanges with it.
These are the commands currently in the MESSAGE package for two protocols, MD and MC (Monte Carlo). New protocols can easily be defined and added to this directory, where LAMMPS acts as either the client or server.
- fix client md = LAMMPS is a client for running MD
- server md = LAMMPS is a server for computing MD forces
- server mc = LAMMPS is a server for computing a Monte Carlo energy
The server doc files give details of the message protocols for data that is exchanged between the client and server.
These example directories illustrate how to use LAMMPS as either a client or server code:
The examples/message dir couples a client instance of LAMMPS to a server instance of LAMMPS.
The lammps_mc dir shows how to couple LAMMPS as a server to a simple Monte Carlo client code as the driver.
The lammps_vasp dir shows how to couple LAMMPS as a client code running MD timestepping to VASP acting as a server providing quantum DFT forces, through a Python wrapper script on VASP.
Here is how to launch a client and server code together for any of the 4 modes of message exchange that the message command and the CSlib support. Here LAMMPS is used as both the client and server code. Another code could be substituted for either.
The examples below show launching both codes from the same window (or batch script), using the “&” character to launch the first code in the background. For all modes except mpi/one, you could also launch the codes in separate windows on your desktop machine. It does not matter whether you launch the client or server first.
In these examples either code can be run on one or more processors. If running in a non-MPI mode (file or zmq) you can launch a code on a single processor without using mpirun.
IMPORTANT: If you run in mpi/two mode, you must launch both codes via mpirun, even if one or both of them runs on a single processor. This is so that MPI can figure out how to connect both MPI processes together to exchange MPI messages between them.
For message exchange in file, zmq, or mpi/two modes:
% mpirun -np 1 lmp_mpi -log log.client < in.client & % mpirun -np 2 lmp_mpi -log log.server < in.server % mpirun -np 4 lmp_mpi -log log.client < in.client & % mpirun -np 1 lmp_mpi -log log.server < in.server % mpirun -np 2 lmp_mpi -log log.client < in.client & % mpirun -np 4 lmp_mpi -log log.server < in.server
For message exchange in mpi/one mode:
Launch both codes in a single mpirun command:
mpirun -np 2 lmp_mpi -mpicolor 0 -in in.message.client -log log.client : -np 4 lmp_mpi -mpicolor 1 -in in.message.server -log log.server
The two -np values determine how many procs the client and the server run on.
A LAMMPS executable run in this manner must use the -mpicolor color command-line option as their its option, where color is an integer label that will be used to distinguish one executable from another in the multiple executables that the mpirun command launches. In this example the client was colored with a 0, and the server with a 1.