ACCELERATION METHODS FOR INPUT DATA CALCULATION UNDER THE 3D-RND PROGRAM COMPLEX AT THE PARALLEL NUMERICAL SIMULATION OF MATHEMATICAL PHYSICS PROBLEMS
K. K. Olesnitskaya, A. L. Potekhin, E. V. Potekhina, V. I. Tarasov
VANT. Ser.: Mat. Mod. Fiz. Proc 2015. Вып.2. С. 49-58.
The paper describes the phases and methods of the parallel calculation efficiency improvement under the 3D-RND complex for the input data calculation of the mathematical physics problems in the 3D setting. In the first method we use the computing load balancing basing on the estimation of the grid cell complexity, according to which they are distributed over the processors. In the second method the whole scope of the сells is divided into big amount of portions (blocks) and their dynamic distribution between the processors. To preserve the computed data in the 3D-RND program complex the EFR library is used, which provides the effective access to the file data on the multi-processor computers. We describe, for the methods, the basic algorithms of calculation and input data preservation, the comparative analysis of the input data calculation rate parameters for a number of problems. Keywords: 3D-RND program complex, EFR library, {it parallel} computations, matrix decomposition, computation load, computation load balancing, computation efficiency.
|
English
