Grand challenge computing: Status and prospects for climate modeling
A climate modeling project using the Intel Paragon reveals the state of parallel computing and identifies issues yet to be resolved. A brief overview of a parallel version of the NCAR Community Climate Model, CCM2, implemented for MIMD parallel computers using a message-passing programming paradigm is described. Performance numbers on computationally intensive kernels indicate that the i860 compilers are doing a reasonable job of optimizing code. Coarse grained decompositions using around 100 processors show good parallel performance while the use of a larger number of processors is impeded by low communication speeds and load imbalance. Climate models generate large quantities of data, some as temporary files and some to be archived. The need for high I/O speeds during the execution of a climate model can be partially offset by a large local memory on each node. Other issues remain, however, for the production use of climate models in a highly parallel computing environment.
- Research Organization:
- Oak Ridge National Lab., TN (United States)
- Sponsoring Organization:
- USDOE, Washington, DC (United States)
- DOE Contract Number:
- AC05-84OR21400
- OSTI ID:
- 50929
- Report Number(s):
- CONF-9303337--1; ON: DE95010216
- Country of Publication:
- United States
- Language:
- English
Similar Records
Parallel community climate model: Description and user`s guide
Computational load in model physics of the parallel NCAR community climate model