Vectorization of a thermal-hydraulics code on the CRAY X-MP

Conference · Thu Jan 01 04:00:00 EST 1987 · Trans. Am. Nucl. Soc.; (United States)

OSTI ID:6861218

Sills, E D; Doster, J M

A thermal-hydraulics code named FLOC, developed specifically to investigate parallel numerical algorithms for solution of the area-averaged Navier-Stokes equations, was originally written to run on a VAX minicomputer and later implemented on a CRAY X-MP. The FLOC code utilizes a semiimplicit formulation to produce a linear system in the spatial pressure distribution at each time step. As transported, the code ran at just over 4 MFLOPs and required 25 CPUs to complete a 10-s pump startup transient for a 72-volume/84-junction two-loop representation of a pressurized water reactor. After restructuring to enhance vectorization by the CRAY vectorizing compiler (version 1.15), the same transient ran at > 23 MFLOPs in < 4 CPUs using fixed time steps. The same simulation allowing variable time steps ran in 0.7 s, more than 10 times faster than real time. The FLOC code consists of eight subroutines, which are called at each time step, and several additional subroutines, which perform initialization and input/output operations. The code allows several choices of direct and iterative algorithms for the matrix solution. The authors have previously reported the preference for direct solution techniques, and the results reported here use the TR algorithm for matrix solutions.

• OSTI ID: 6861218
• Report Number(s): CONF-8711195-
• Conference Information: Journal Name: Trans. Am. Nucl. Soc.; (United States) Journal Volume: 55
• Country of Publication: United States
• Language: English

Related Subjects

42 ENGINEERING
420400* -- Engineering-- Heat Transfer &amp; Fluid Flow
ALGORITHMS
COMPUTER CODES
COMPUTERS
CRAY COMPUTERS
DIFFERENTIAL EQUATIONS
ENERGY TRANSFER
EQUATIONS
F CODES
FLUID MECHANICS
HEAT TRANSFER
HYDRAULICS
MATHEMATICAL LOGIC
MECHANICS
NAVIER-STOKES EQUATIONS
NUMERICAL SOLUTION
PARALLEL PROCESSING
PARTIAL DIFFERENTIAL EQUATIONS
PROGRAMMING