Project Description
This section will cover the business issues, product statement and projected
value of the PPF project. All estimates and concerens are preliminary and will
undoubtadly change as the project progresses.
Business Issues
Radiation particle packages are written largely by physicists without formal backgrounds in computer
science. Learning the intricacies of distributed computing and applying
it to a project
increases the time of development and may compromise the correctness of that project. More
specifically, the business issues surrounding the parallelization of radiation transport
packages are as follows:
- Parallel training increases project times.
- Parallel computation is not germain to many package developers fields
and can be viewed as a waste of resources.
- Efficiency and correctness of parallelization requires experience and knowledge
of architectures that many radiation transport developers lack.
- Currently, each developer is approaching the parallelization problem differently. The absence of a common structure makes this class of code difficult to maintain
- Current parallelization techniques used are not conducive to future changes, such as:
- increased scalability
- load balancing
- optimization techniques
- Current parallelizations are not abstracted from the code to support adaption to new
parallelization technology.
Product Statement
The Parallel Particle Framework (PPF) project involves designingand developing an API which will be integrated into physics modules
within the KULL framework. The API will provide a high level interface for the parallelization
of particles over large meshes\footnote{a mesh is a collection of
geometric components which represent a problem domain}
for radiation transport physics packages developed for KULL.
The PPF will decompose the problem mesh over large distributed
super computers, and handle all communication of particles moving between
nodes. The API must be constructed to integrate into existing physics packages
without changing the original programs' paradigms.
Projected Value
The goals of the PPF are to save time in developing particle Monte Carlo packages
and to provide a standard method of parallelization. Benefits of a technological solution are:
- Reduction of development and training hours in order to reduce the time and
cost of a project.
- Maximization of resources by enabling physicists to concentrate
on work applicable
to their areas of study.
- Ensured correctness in parallelization.
- Increased maintainability of code, due to standardization.
- Ensured robust, efficient parallelization over distributed systems, improving the end
product.
- Preparation for future changes to distributed mechanisms by abstracting parallelization
methods.
Business Environment
The parallel particle framework is intended to be integrated into radiation
transport packages within the KULL framework. The KULL project environment
requires the PPF to adhere to the following constraints:
- It must compile on Digital Unix and AIX using the KCC compiler.
- It must conform to the C++ standard.
- It will follow the KULL documentation standards.
- Builds should be easily integrated into the larger KULL make system.
Since the PPF is meant to be integrated not only into new projects but
existing code, the interface must be conducive to the parallelization
of
serial code. That is to say, PPF must not require the project architecture
to be built around it, but more so be easily integrated into existing
architectures.
In general, problems which use the PPF must accept following assumptions:\\
- Particles do not interact with each other. They are absolutely
independent.
- Particles always contain the following information:
- their position in the Mesh( x,y,z)
- their direction of travel( u,v,w)
- time( t) since being sourced
- energy E carried
- mesh-zone they are currently in
- The mesh is rigid and particles move through the mesh.
- Particles change mesh based values(heat, energy, etc.) which are the
answer to the problem.
[HOME]