The ability for heterogeneous distributed computers to interoperate seamlessly is a core requirement for the DOE 2000 effort. The above graphic illustrates many of the exciting areas of distributed computing research at ORNL that support this effort.
ORNL has developed a world-wide reputation for its parallel and distributed computing research. The PVM software developed at ORNL has become the de facto standard for distributed computing in businesses, universities, and laboratories. ORNL also played a key role in creating the Message Passing Interface (MPI) standard and its follow-on MPI-2. ORNL has one of the most powerful parallel computers in the world in its Center for Computational Science. and is also a partner in the national High Performance Storage System project.
This expertise is now being channeled into the development of a "Seamless Distributed Computing Environment" where scientists and engineers can collaborate and exploit the aggregate power of heterogeneous supercomputers connected all across the country by high speed networks. The goal of the environment is to promote collaboration through ease of access and sharing of computational and scientific resources. The vision of the Seamless Distributed Computing Environment is to allow multiple sites to participate in computational experiments as well as physical experiments regardless of the location of the physical hardware.
The Seamless Distributed Computing Environment
is composed of several interoperating sub-environments
each of which is a major software development effort.
Uniform Program Development Environment
- includes debuggers, compilers, profilers, editors, msg passing, which all interoperate with each other no matter what site a user is at.
Seamless Runtime Environment
- includes schedulers, common file system, I/O issues, security, resource management, visualization, and collaboration tools, and real-time remote control.
Problem Solving Toolkit
- includes math libs, interoperating modules for PDE, grid generation, partitioning, etc., fault tolerance.
The development of these sub-environments requires the coordinated efforts of multiple organizations and DOE labs. As a start, this past year ORNL and Sandia National Laboratory began a collaborative research project to hook together the two biggest Intel Paragons in the world using high speed ATM OC12 network interfaces. In order to utilize this distributed computational resource, both labs are also collaborating on the development of prototype software for the Seamless Distributed Computing Environment.
Highlights from ORNL's parallel and distributed computing research this past year include: