The MapReduce model can be divided into single-reduce mode, multi-reduce mode and no-reduce mode, and different mapreduce calculation mode should be chosen according to the demand for the production algorithm of the exponential product with different complexity.
1) Low complexity of product production algorithms
for the low complexity of the production algorithm of remote sensing products, generally only need to use a MapReduce compute task, you should choose the multi-reduce mode or no reduce mode.
When the index product algorithm involves only one file in the input data (such as the production of a landscape global Environmental monitoring index products, only one landscape HDF format of the MODIS land level two product data), you can choose No reduce mode . The map phase is responsible for implementing the core algorithm of the index product. Specific calculation processes such as:
when the index product algorithm involves the input data containing multiple files (such as the production of a Prairie drought index products, need to use the surface reflectance, surface temperature, rainfall, etc.), you should select the multi-reduce mode. The map phase is responsible for collating the input data, and the reduce phase is responsible for implementing the core algorithm of the index product. Specific calculation processes such as:
2) Product production algorithm with high complexity
for the high complexity of remote sensing product production algorithm, a MapReduce computing task is often difficult to meet the production requirements, you need to use multiple mapreduce tasks to accomplish the production tasks of the product together. In this case, you can solve the dependencies between tasks by using the Oozie workflow engine to control the workflow of multiple MapReduce compute tasks. Oozie Introduction and Installation tutorial can refer to another blog post: http://blog.csdn.net/until_v/article/details/40682205
How to use Hadoop to realize different complexity of remote sensing product algorithm