Abstract:

The size of climate model output continues to grow and is overwhelming the tools scientists use to postprocess the data. The biggest reason current tools cannot keep up is that they are all single-processor programs. The Parallel Climate Analysis Library, ParCAL, is a new set of data-parallel functions designed to vastly improve the post-processing or analysis time of the ultra-large climate data compared to the current serial tools. Typical climate analysis functionality such as that provided by NCAR Command Language (NCL) are being implemented, where possible, with advanced algorithms from other parallel libraries or by writing our own. As ParCAL is being developed, it is making use of several software engineering technologies. SVN is our choice of version control system. The build system fully relies on autotools (autoconf, automake and libtool), which enables automatic generation of all configuration scripts required. The Boost Test Library - Unit Test Framework provides a set of convenient functionalities for performing unit tests. All of the algorithms implemented so far are fully tested with the Boost Test Library. Nightly build/test is employing the buildbot system to enable the automation of the build and test cycle and also provide a web interface for browsing and monitoring the test processes. ParCAL is tested every night with various configurations, such as the trunk test, documentation generation (with Doxygen) and different compiler choices etc. We are using a Trac-based ticket system for issue and bug tracking.