How to define your simulation (single bunch)

To define a simulation to run with the PyPARIS ring of CPUs, the user needs to write a simulation class. Each CPU will instantiate a separate instance of the class, and use it to preform the simulation according to the paradigm defined here.

PyPARIS will add to the object a member called ring_of_CPUs, which carries the information about the multiprocess structure and allows each process to identify itself.

A prototype simulation class is defined in the following, and examples can be found in the PyPARIS test folders.

class Simulation(object):
	def __init__(self):
		self.N_turns = 128
		self.N_buffer_float_size = 500000
		self.N_buffer_int_size = 450

	def init_all(self):
		# This function is executed by all nodes at the beginning of the simulation 
		# and it should be used to initialise the simulation.
		# All the relevant data and code should be stored as members of this 
		# simulation object. 
		# Each CPU can identify itself and get information about the ring of CPUs
		# through the object self.ring_of_CPUs that is automatically attached by 
		# PyPARIS (e.g. self.ring_of_CPUs.myid, self.ring_of_CPUs.N_nodes,  
		# self.ring_of_CPUs.I_am_the_master, self.ring_of_CPUs.I_am_a_worker).

	def init_master(self):
		# This function is executed by the master process only, after the  
		# init_all function. It must return a list of pieces object (e.g. bunch
		# slices) to be fed to the ring of CPUs at the first turn.
		return pieces_to_be_treated

	def init_worker(self):
		# This function is executed by all the worker processes (and not by the  
		# master), after the init_all function.

	def treat_piece(self, piece):
		# This function defines the task to be performed by each process when it 
		# receives a piece. (Most of the computational time of the simulation is 
		# most likely spent here).

	def finalize_turn_on_master(self, pieces_treated):
		# This function is executed on the master at the end of each turn, i.e. 
		# when all pieces have gone through the complete ring of CPUs and have 
		# been recollected by the master process. 
		# At this point you can re-merge the pieces on the master process (e.g. 
		# merge your slices) and perform operations that need to be performed 
		# collectively on your set of pieces (e.g. synchrotron motion).
		# Then you can generate a new set of pieces to be treated (e.g. re-slice
		# your bunch), which needs to be returned by this function.
		# Through this function you can broadcast messages ("orders") from the 
		# master to all the other processes. These messages need to be in the
		# form of a list of strings and will be passed to the function 
		# execute_orders_from_master, which is executed by all the processes
		# (including the master).

		return orders_to_pass, new_pieces_to_be_treated


	def execute_orders_from_master(self, orders_from_master):
		# This function is executed by all processes at the end of each turn and 
		# is supposed to execute the orders broadcasted by the master (see 
		# function: finalize_turn_on_master).


	def finalize_simulation(self):
		# This function is executed by the master process at the end of the 
		# simulation. It can be used to save the results of the simulation or 
		# for data post-processing.
			

	def piece_to_buffer(self, piece):
		# This function will be used to covert a piece object into a buffer (in 
		# the form of an array of floats), which is then sent to the 
		# destination process.
 		# In the case in which the object is a PyHEADTAIL beam, the converter
 		# is provided within PyPARIS in the communication_helpers module.
		return buf

	def buffer_to_piece(self, buf):
		# This function will be used to covert a buffer (in 
		# the form of an array of floats) sent by another process into a piece 
		# object to be used in the computation.
 		# In the case in which the object is a PyHEADTAIL beam, the converter
 		# is provided within PyPARIS in the communication_helpers module.

		return piece