Author/Authors :
K.-Y. Wanga، نويسنده , , 1، نويسنده , , )، نويسنده , , D.E. Shallcrossb، نويسنده , , S.M. Hallc، نويسنده , , 1، نويسنده , , Y.-H. Loa، نويسنده , , 1، نويسنده , , C. Choua، نويسنده , , 1، نويسنده , , D. Chena، نويسنده , , 1، نويسنده ,
Abstract :
With the increasing computational speed of PC processors, lowering hardware prices, and the increasing accessibility of open
source software, PC clusters have become an attractive option for exploring budgetary high performance computation on high
resolution environment modelling. This paper documents the implementation and operation of a Beowulf-class PC cluster called
DOBSON built in the Atmospheric Modelling Laboratory at National Central University in Taiwan. Based on our current
configuration, the DOBSON cluster achieves a 4.4–8.7 times of speedup with 8 and 14 CPUs, respectively, compared with a 1 CPU
in simulations of the Antarctic vortex. The potential impacts of running a high resolution environmental modelling system on the
DOBSON cluster were examined by case simulations of Typhoon Haiyan with a nested five-domain configuration. These results
show that the complicated topographical flow in the Taipei Basin can only be properly resolved when model grid resolutions less
than 9 km were used.
Keywords :
Air pollution modelling , Cluster computing , Distributed-memory computing , Beowulf