Research on parallel computing architecture of meteorological early warning based on supercomputing platform

Author's brief introduction ： Wuhan Mavericks focus on GIS Big data applications 、 Microservices 、WEBGIS

【 Application scenarios / The main solution is 】

        In the future, the accuracy of meteorological early warning is required to be higher and higher , Not only to increase the density of meteorological sampling points , It also requires more detailed forecast service products , Information release is faster . The supercomputing platform is mainly used to improve the release speed of provincial meteorological early warning calculation , Achieve minute level warning , Avoid and reduce casualties and economic losses caused by geological disasters to the greatest extent .

【 Content description 】

The resource configuration of the computing node is as follows ：

in the past , The software is written in serial , Suppose in a CPU Above to perform the task . Supercomputers have completely broken this situation , Turn to emphasize “ Parallel computing ”. The principle of high-performance computing is not complicated , Is to divide a task into several parts , Each server completes one , But the basis of parallelization is a thorough understanding of the application business involved in this code , On the basis of understanding , Then carry out reasonable task segmentation , Handle load balancing finely 、 Communication, etc .

Let's take the meteorological early warning calculation model as an example , From functional parallelism to data level parallelism 2 The specific design is elaborated in three aspects , The overall architecture is as follows ：

The computational unit analysis step is designed as a pipeline job that can run independently , Having independent data is divided by administrative region . Each unit calculation is completed , Send message notification to controller , Calculate the completion progress of the whole task . Graphics real-time data is stored in memory , By extending the GEO SERVER To support the publishing of dynamic graphs .