High Tide in Ocean Modeling

The Pittsburgh Supercomputing Center (PSC) is committed to advancing the state of the art in scientific computation. One cost-effective way of doing this is to encourage the development of new software which runs efficiently on massively parallel processors. One notable example of scientific accomplishment using these techniques is ocean modeling by computer engineers Matt O'Keefe and Aaron Sawdey of the University of Minnesota.

The goal of this project is to simulate circulation in the the Atlantic Ocean using the Miami Isopycnic Coordinate Ocean Model (MICOM), developed by University of Miami ocean scientist Rainer Bleck and his colleagues. In recent computations at PSC, in a very high-resolution simulation that ran for 10 days on 256 T3D processors (see below), MICOM correctly predicted the course of the Gulf Stream. No other model of the entire Atlantic Ocean has achieved this degree of realism.

Surface Temperature of the Atlantic Ocean
As the results show (red corresponding to high temperature), the Gulf Stream follows a realistic course, breaking off from Cape Hatteras and proceeding east-to-northeast into the open Atlantic.

In prior ocean models of this scale, the Gulf Stream stayed close to the continental shoreline north of Cape Hatteras rather than taking to the open sea. Researchers suspected grid-size was the problem, and the T3D provided an opportunity to test this surmise. With resolution of 0.08 degrees longitude (roughly six kilometers at mid-latitude), the Gulf Stream immediately established itself on the right course. This is an important result for ocean modeling, placing MICOM at the forefront of the field. It represents a notable step in the direction of being able to provide reliable predictions of ocean levels related to concerns about global warming.

Parallel Algorithm Development at the PSC

As a National Science Foundation sponsored high-performance computing center , a foremost goal of the Pittsburgh Supercomputing Center is to make the most powerful high-performance computing resources available to the nation's scientific researchers. In accordance with that goal and with additional financial support from the Advanced Research Projects Agency ARPA and the National Institutes of Health NIH, PSC accepted delivery of the very first T3D Massively Parallel Processor (MPP) from Cray Research in August of 1993. In its current configuration the T3D has 512 processing elements each of which is a DEC Alpha processor with 64 Megabytes of memory. These processors are connected by a very high speed network organized as a three-dimensional torus. The maximum theoretical speed of the T3D is about 75 Gflops (75 billion floating point operations per second). Another important feature of the T3D is its ability to combine with the PSC's Cray C90 pipelined vector processor for heterogeneous processing; in this mode each supercomputer can work on that part of a computational problem which is most well suited to its architecture.

The major impediment to the use of such massively parallel computers is the relative lack of applications software which can exploit the power of the hardware. To vigorously address this problem the PSC and Cray Research (and some other high-performance computing centers) have combined their efforts as part of a project called the Parallel Applications Technology Program (PATP). This work has already enabled important scientific results in materials science, protein structure, DNA structure, ocean modeling, organization of the visual cortex, chemical clusters, and quantum chromodynamics. Details of some of these projects are available in the PATP Annual Report. PSC staff working with Cray applications programmers have also made available on the T3D a set of third party codes widely used in the scientific community. These include Amber, GAMESS, CHARMM, Shake'n'Bake and MaxSegs. Work is underway on a number of additional packages including X-PLOR and FLAPW.