Numerical Weather Prediction on the Supercomputer Toolkit

  • Authors:
  • Pinhas Alpert;Alexander Goikhman;Jacob Katzenelson;Marina Tsidulko

  • Affiliations:
  • -;-;-;-

  • Venue:
  • ISHPC '02 Proceedings of the 4th International Symposium on High Performance Computing
  • Year:
  • 2002

Quantified Score

Hi-index 0.00

Visualization

Abstract

The Supercomputer Toolkit constructs parallel computation networks by connecting processor modules. These connections are set by the user prior to a run and are static during the run. The Technion's Toolkit prototype was used to run a simplified version of the PSU/NCAR MM5 mesoscale model [9]. Each processor is assigned columns of the grid points of a square in the (x,y) space. When n 脳 n columns are assigned to each processor its computation time is proportional to n2 and its communication time to n. Since the Toolkit's network computes in parallel and communicates in parallel, then, for a given n, the total time is independent of the size of the two dimensional array or the area over which the weather prediction takes place. A mesoscale forecast over the eastern Mediterranean was run and measured; it suggests that were the Toolkit constructed from ALPHA processors, 10 processors would do a 36 h prediction in only about 13 minutes. A 36 hours prediction with full physics for the whole earth will require 2 hours for 80 ALPHA processors.