Global Index (all files) (short | long) | Local contents | Local Index (files in subdir) (short | long)
[OP1, OP2, OP3] = dpgeamain(StrTask, P1, P2, varargin)
DPGEA Matlab: Main function
This function is the main function to use the DP-Toolbox functionality
inside own projects.
This function includes a number of subfunctions to achieve the
following tasks:
- start the PVM system
- start slaves (Matlab instances)
- distribute tasks to slaves and collect the results
- check running instances
- kill slaves
- clear left over data from buffers
- get information about hosts (architecture, name and so on)
By using this function a high level interface to the distribution of
tasks inside Matlab is given. The user must include just a few lines
of code in own projects.
The high-level interface hides many of the complications of the
DP-Toolbox and PVM, offering just the functionality needed for solving
the problem.
Syntax: [OP1, OP2, OP3] = dpgeamain(StrTask, P1, P2, varargin)
Input parameter:
StrTask - String defining the task to do
P1 - Px - Input parameters for the different tasks
see description of tasks for their meaning
Output parameter:
OP1 - OPy - Output parameters for the different tasks
see description of tasks for their meaning
Task descriptions (string inside StrTask):
'start' - start PVM (if not already running) and spawn the slave
proceses inside MATLAB instances,
instances can be spawned on specified machines (defined
by the parameter P1) or on default machines (determined
by PVM), the spawned slaves are added to the current
configuration
P1: cell array with strings containing the names of the
machines to start instances on or
with a number, how many slaves to start
no output parameters
Example: start 3 slaves, one on bfo-galilei, two on bfo-kalman
>> dpgeamain('start', {'bfo-galilei', 'bfo-kalman','bfo-kalman'})
'master' - distribute the received data to the Matlab instances (slaves),
do some robust test and do task balancing between clients,
receive the results from the slaves
distribution is done one task after another
P1: string containing the function name to call with the data
(often called objective function)
P2: matrix with data to distribute to function in P1, each
row contains one data set
varargin: (optional) cell array with additional parameters to
send to function defined in P1, each cell element contains
one parameter (each parameter could be a scalar, a vector or
a matrix or a cell array ...)
OP1: vector or matrix with results, each row corresponds to
one row of the input data defined in P2
Example: distribute 10 tasks to the function 'objfun1' (slaves must
be started before)
>> data = rand(10, 100); % 10 tasks, each with 100 variables
>> Results = dpgeamain('master', 'objfun1', data);
distribute 1000 tasks to the function 'objdopi',
send 2 additional parameters
>> data = rand(1000, 20); % 1000 tasks, each with 20 variables
>> Tstart = 0; TEnd = 2; % define additional parameters for 'objdopi'
>> Results = dpgeamain('master', 'objdopi', data, {TStart, TEnd'});
% 'objdopi'will be called from dpgeaslave with the following line
% xxx stays for the number of the current task
% >> Result_for_data_xxx = objdopi(data(xxx,:), TStart, TEnd);
'check' - checks, if pvmd and/or matlab is answering
P1: == 'pvmd': checks just the pvmd of the given ID's
== 'matlab': checks the pvmd and calls the (not working)
matlab instances
P2: vector containing the number (PVM ID) of the instances
to check,
OP1: vector containing the IDs of the instances, that are ok
OP2: vector containing the IDs of the instances, that are not ok
Example: check, if the pvm daemons of slaves are responding
>> [IDs_ok, IDs_bad] = dpgeamain('check', 'pvmd', IDs_2_check);
'clear' - clear the recive buffer from arrays, used when suspending
some of the Matlab instances
no input or output parameters
'kill' - kill/stop all or specified running MATLAB instances
and adjust global variables
P1: vector containing the number (PVM ID) of the instances
to kill,
if omitted, all running instances are killed
no output parameters
Example: kill all running slaves
>> dpgeamain('kill')
'hostname'- look up information (hostname, architecture, speed) of
multiple instances
P1: vector containing the number (PVM ID) of the instances
to get information for,
if omitted, the information for all started tasks is returned
OP1: string array containing the hostname information
OP2: string array containing the architecture information
OP3: string array containing the speed information
Example: get host name and architecture of ID1 and ID2
>> [hn, ar] = dpgeamain('hostname', [ID1, ID2]);
See also: dpgeaslave
| This function calls | |
|---|---|