CCSM Software Engineering Working Group Meeting
24 June 2003
The CCSM Software Engineering Working Group (SEWG) met during the CCSM Workshop, on June 24, 2003, from 1:00 p.m. to 5:00 p.m. The following presentations were given during that meeting.
Tony Craig talked about the status of CCSM, release plans, platforms, vectorization, infrastructure, SE process, and priorities. CCSM2.2 is expected to be released at the end of 2003. CCSM currently runs on IBM, SGI Origin, and HP/Compaq. Work is in progress to port CCSM to Linux clusters, the Cray X1, and the NEC SX6. CCSM has been run successfully on a few Linux clusters including jazz at ORNL. There are still plans to upgrade the CVS server and CVS scripts. There is an effort within CSEG to improve testing and the testing process. A new suite of scripts and a new GUI will hopefully be released with CCSM at the end of the year.
John Drake provided a summary of SciDAC activities.
Craig Rasmussen presented a comparison of the Earth System Modeling Framework (ESMF) and the Common Component Architecture (CCA) designs. He showed that the goals of the two component frameworks are different. The ESMF provides a specific design for building component applications for the climate community, while the CCA provides a general specification for building general components that can be assembled at runtime. With very minor modifications, ESMF components can be extended to be CCA components. This allows ESMF applications to be composed at runtime and also allows ESMF components to connect to general CCA components, thus extending the capabilities of an ESMF application.
Cecelia DeLuca gave an update on the status of the ESMF project. She briefly discussed the component interface to ESMF and some of the main ESMF capabilities.
Erik Kluzek presented the work going on to implement ESMF into CAM and CCSM. He presented efforts at pseudocoding what CAM and CCSM will look like with ESMF, as well as the implementation plan and strategy. Erik showed that subroutinization of CAM is an important step of adopting ESMF for CAM, but also allows isolation of physics from dynamics in CAM. This makes it easier to interchange new dynamics or physics into CAM, and eliminates duplicated code. Thus, in the long run, implementing ESMF into CCSM will cut down on our software engineering costs for CAM as well as the other CCSM models.
Rob Jacob presented work done to instrument CCSM2.1beta with the MPE profiling library. Custom MPE timers were created and introduced into the code to distinguish between coupling and computation phases of each model-processor. Several configurations of CCSM, with different processor counts for the components, were integrated on bluesky with a goal of increasing the total throughput. The results show that data dependencies between the atmosphere, land, and ice models lead to load imbalance within the coupled system. The scaling of the full system is currently limited by the scaling of the ocean component. A different selection of runtime options may improve the latter.
Mariana Vertenstein presented work being done on the new CCSM GUI.
V. Balaji provided an overview of the PRISM System: a European software project to build an open standard coupling interface for climate model components, as well as a standard user interface for model configuration and analysis.
There were no open discussion items.