SpiNNaker-PACMAN 1!7.1.0

Partition and Configuration Manager

This package provides utilities for partitioning, placing a routing on a SpiNNaker machine

Requirements

In addition to a standard Python installation, this package depends on:

• SpiNNMachine

These requirements can be install using pip:

pip install SpiNNMachine

User Installation

If you want to install for all users, run:

sudo pip install PACMAN

If you want to install only for yourself, run:

pip install PACMAN --user

To install in a virtualenv, with the virtualenv enabled, run:

pip install PACMAN

Using a virtual environment is recommended for all SpiNNaker software.

Developer Installation

If you want to be able to edit the source code, but still have it referenced from other Python modules, you can set the install to be a developer install. In this case, download the source code, and extract it locally, or else clone the git repository:

git clone http://github.com/SpiNNakerManchester/PACMAN.git

To install as a development version which all users will then be able to use, run the following where the code has been extracted:

sudo pip install -e .

To install as a development version for only yourself, run:

pip install -e . --user

To install as a development version in a virtualenv, with the virutalenv enabled, run:

pip install -e .

Test Installation

To be able to run the unitests add [Test] to the pip installs above

pip install -e .[Test]

Documentation

PACMAN python documentation
Combined python documentation

PACMAN Executor

PACMAN contains a simple workflow execution system which allows the user to specify a set of available algorithms which, when provided with a set of inputs can produce a set of outputs. The executor will then work out the order in which the algorithms should be run (and indeed if the algorithms can be run) by looking at the inputs required and outputs generated by each algorithm.

As well as parameters required by the algorithms, the workflow system additionally supports the concept of "tokens". A token can be used to represent the action of an algorithm that does not produce a specific output. For example, on a SpiNNaker machine, this might include the loading of data or the loading of application binaries, neither of which produces a Python object as output, but performs and important task in any case. Each token can also be specified to be "part" of a whole task. This allows an algorithm to declare that it has done part of a task, and have a future algorithm require that all of the task has been completed without knowing what parts need to be done. Again, the example of loading of data can be used here where there may be several algorithms that can load data in different ways but all the data must be loaded before the application binaries are loaded; however the application binary loader can just say that all data loading is done before execution, avoiding the need to modify the algorithm in the event that a new algorithm is created.

The arguments of the algorithms that can be represented as Python objects are specified using semantic types. These are simply represented as strings; the values of the strings are only important in that they must match between the inputs and outputs of algorithms in the flow for an algorithm to be recognised as producing an output that another algorithm can use as an input.

Each algorithm to be run by this executor specifies:

• The required inputs of the algorithm. This can include both arguments of the algorithm and tokens required.
• The optional inputs of the algorithm, again including both arguments and tokens.
• The outputs generated by the algorithm, both as semantic types and tokens.

The executor is provided with:

• A list of algorithms that must be executed. If it is not possible to execute these, an error is raised. These will be executed regardless of if the output of the algorithm is already available.
• A list of algorithms that can be executed optionally to produce an output. If one of these algorithms generates an input type or a partial or complete token that is an optional input to a required algorithm, and no other required algorithm can generate this input, the optional algorithm will be run before that required algorithm. An optional algorithm will not be run if it doesn't generate an output which is an input for another algorithm which isn't already provided in some other way.
• A list of inputs to seed the workflow with. This can be used to provide initial inputs to the workflow, or to stop an optional algorithm from executing by providing in advance the output that the algorithm generates.
• A list of input tokens to seed the workflow with. These can only be specified as complete tokens which are provided. Partial tokens cannot currently be provided.
• A list of required outputs that the workflow must generate. In addition to the list of algorithms to run, these outputs must be generated by the workflow at some stage, or else an error is raised.
• A list of required output tokens that the workflow must generate. As with the outputs these must be generated by the workflow at some stage or else an error is raised.