A priority packet scheduling method and system utilizing data topological information

Abstract

The invention provides a priority packet scheduling method and system utilizing data topological information. The method comprises the steps of obtaining an original grid space of data topological information, setting the size and the floating point precision of a grid sheet of the original grid space and generating a new grid space; according to the stencil format of the new grid space and a parallel zone, establishing a condensed task graph (four-dimensional time-space domain) and calculating a priority packet serial number of each task therein; obtaining a data piece visited by the current task, determining whether neighbor data dependence is involved through format abstract or function pointer and actual parameters , generating a corresponding mark, and identifying a cycle involving the neighbor data dependence according to the mark, wherein the cycle is effective time steps; mapping the task on the cycle to some task of the condensed task graph according to the current effective time steps, calculating the priority serial number of the current task according to the priority serial number of the latter. The priority packet scheduling of tasks is supported.

Description

Technical field [0001] The invention relates to the field of computer and user-level task scheduling, in particular to a priority grouping scheduling method and system using data topology information. Background technique [0002] There is an insurmountable memory wall in the modern microprocessor structure. How to optimize the data locality in the program and improve its cache performance on the processor is an important subject of application optimization. [0003] The background of the structured grid method is as follows: there are a large number of nonlinear equations in the fields of thermal diffusion, electromagnetic field and fluid mechanics, and only a few problems can be solved analytically or perturbed. The structured grid method is to solve such problems. One of the most important numerical methods for the problem, its typical steps are: discretize the solution domain, convert the continuous solution domain into a finite set of discrete points, such as a fixed equidista...

AI Technical Summary

Problems solved by technology

First of all, this type of method is very complicated to implement—it must be based on a mature domain programming framework for application development, the data structure becomes complex and is no longer a simple array, and the calculation steps are also mixed with C / Fortran, which leads to Compilation is difficult to perform static dependency analysis on the entire grid space and each calculation step, so timeskewing cannot be realized

Second, due to the dynamic nature of grid division, the neighbors of grid points cannot be determined statically, which also limits the implementation of timeskewing optimization

) No one has found that priority group scheduling can be used to improve task scheduling performance in structured grid applications

Images