Client-Server Operation
Qbox can be used in client-server mode. This means that Qbox can run on one computer as a server that “listens” to commands being sent from another computer. Commands can be sent interactively by a user, but also by another program (the “client”, or “driver”). It is assumed that the computers used by the client and the server share a common file system. Typically, the client and server are different processes running on the same computer.
Client-server operation allows for extensions of the functionality of Qbox in which the client and the server interact to achieve a result that is not easy to get with the conventional approach of submitting a fixed input file to Qbox. A Qbox driver (or “Qbox application”) program sends commands to Qbox by writing them to a file. Qbox reads the commands, executes them, and produces an output file. The client code can then read the output file, process it, and send the next commands to Qbox. Synchronization of this process is implemented through the creation (or destruction) of a file (the “lock” file) that signals that the output (or input) file is ready for use. Qbox is run in client-server mode by invoking it as follows:
$ mpirun -np <ntasks> qbox_exec -server input_file output_file
The sequence of operations in client-server mode is described below
Qbox:
Create a new file named output_file.
Read commands from input_file and execute them until the end of file is reached. Write the output on output_file. After executing the last command in input_file, close output_file and create a file named input_file.lock to signal that Qbox is ready for more commands.
Wait for the file input_file.lock to be removed by the driver.
Go to 1)
Client:
Wait for the file input_file.lock to appear.
Open the file output_file and read its contents.
Analyze the contents and decide what course of action to take.
Open the file input_file and overwrite it with new commands.
Close input_file.
Remove input_file.lock to signal that the file input_file is ready.
Go to 1)
Note that in server mode, Qbox only exits when it reads the quit command. When reaching the end of file in input, Qbox does not exit but creates the lock file and then waits for the client to remove it. It then starts reading the input file again.
The contents of output_file form a valid XML document, i.e. The XML header is repeated every time the output file is rewritten. This allows one to process each output file using XML tools, such as e.g. XSLT scripts or a python XML parser.
A client can manage multiple copies of Qbox running in server mode. In that case, multiple copies of Qbox must be launched with different input and output file names. For example:
$ mpirun -np <ntasks> qbox_exec -server in0 out0 &
$ mpirun -np <ntasks> qbox_exec -server in1 out1 &
$ mpirun -np <ntasks> qbox_exec -server in2 out2 &
The client must then write on the files in0, in1, in2, and read output from the files out0, out1, out2. This allows for simulations involving multiple samples, such as replica exchange dynamics, nudged-elastic-band (NEB) simulations, or path-integral molecular dynamics.