Distributed supercomputing for climate modeling?

Many scientific applications are of such complexity and scale that solutions can be obtained only by using a wide variety of computing hardware – all at once. The need for concurrent use of e.g. multiple clusters, grids, clouds, and supercomputers has spawned much innovative research, and user-oriented programming models for such advanced systems are under constant development. Examples of applications that benefit particularly well from these advances are so-called multi-model/multi-kernel simulations, in which multiple distinct models of real-world phenomena are coupled to form a single, large simulation of a physical system. Such simulations are common in climate research, where models of land, ocean, atmosphere, and ice are combined to simulate the earth’s climate.

In this presentation we describe an experiment that attempted to run a single climate simulation simultaneously on four different supercomputers located in The Netherlands, UK, Germany, and USA (including PRACE and XSEDE resources). These supercomputers where interconnected by state-of-the-art 10G networking technology (lightpaths) provided by various NRENs as part of the Enlighten Your Research global (EYRg) competition. Although some results where obtained, the experiment was not entirely succesful. In the presenation we will describe the problems we encountered in more detail, and provide suggestions on how these problems may be avoided in the future.