|
The Chiba City Project
Chiba City is a scalability testbed for the High Performance Computing and
Computer Science communities. It is designed to be a tool to help answer
questions such as:
- What issues arise when today's existing libraries and code scale to
hundreds or thousands of systems?
- What algorithms can scientific code use to scale to high performance
computing systems of the near future, when systems will comprise thousands
of nodes?
- What types of systems software and systems management tools can be
built to make these large-scale systems operate efficiently?
- What new approaches can be considered to tackle problems of large
scale?
- How can we effectively provide large-scale testbeds to the computer
science community?
- Can a system be built of commodity components and open source tools
that virtualizes the functions of a large computing resource?
The Mission
Research projects on Chiba are concentrating in four key areas:
- Scalable Libraries and Middleware
- Scientific Vizualization
- Distributed Computing
- Systems Software and Cluster Management
In addition, Chiba is a testbed for open source
development of system software for high performance computing.
The Cluster
The Chiba City cluster is a 512 CPU cluster primarily running Linux. The
cluster also includes a set of storage nodes, visualization nodes, and
server nodes, as depicted in the
configuration diagram.
The cluster was built in partnership with
IBM and
VA Linux Systems.
Availability and Usage
Chiba City is a system designed to support research and development. It
is primarily focused on computer science scalability research. The system
is continually under development, and is not meant to provide
production computing cycles for number-crunching.
Chiba City is available on a limited basis for computational scientists who
are collaborating with MCS scientists, and more generally available to
computer scientists who are doing research in scalability. For
information on requesting an account and using the system, please see
the links along the left side of this page.
|