Oxford University, the Met Office and several other
universities have developed a huge grid computer for the BBC to
predict climate change up to 2080.
The climateprediction.net experiment harnesses the spare
computing power of tens of thousands of PCs, and will rival
supercomputers for power, said Oxford University.
The system will rank among other massive public grids such as
the search for extra-terrestrial intelligence (SETI) and the human
genome project.
The BBC has invited volunteers to download an application, which
is a version of the Met Office's state-of-the-art climate model. It
processes data automatically when their PC is online, and sends it
back to Oxford University.
"In certain applications, such as our massive ensemble of
climate simulation, we are achieving greater than supercomputer
levels of computing power," said Carl Christensen, chief software
architect of climateprediction.net project at Oxford
University.
But he added that the grid will not replace supercomputers,
which are still required for very high resolution climate and
weather modelling, due to high memory and CPU usage, and the volume
of data output.
Surprisingly, the grid is co-ordinated using "a few
off-the-shelf servers", one for the database and one for the
website, plus a few desktop PCs, said Christensen.
The servers are 2.4GHz dual-Intel Xeon systems with 2Gbytes of
memory, but can support data from 80,000 volunteers.
The project uses very fast C++ client and server code, and the
mySQL database, a system pioneered by the University of California,
Berkeley.
"Even a humble server set-up such as ours can handle quite a
large number of user requests, so basically for a few thousand
dollars you can support a project that is doing 100 teraflops," he
said.
The project was not without its difficulties, though.
"The primary technical challenge has always been in porting the
climate model, which is about a million lines of legacy Fortran
code developed by hundreds of scientists over the past two
decades," said Christensen.
Developers had to tailor this for various PC platforms,
including Windows, Linux and Mac OS X.
Another challenge is the small number of computing employees on
the project - there are only two full-time computing staff carrying
out system administration, code development, database programming
and managing the website and graphics.
Also taking part in the project are the University of
California, Berkeley; the University of Reading; Open University;
and Rutherford Appleton Laboratory.