The observational data partitioning scheme has been encapsulated into a separate module called OBSORT. The program OBSORT redistributes the observational data across the available processors. Data supported must be either in the CMA and/or BUFR formats. As a result the subsequent steps in the general analysis data flow will be well load-balanced with respect to observation handling and the total elapsed time for the analysis is reduced. The facilities offered by the OBSORT can be used as a stand-alone executable, or used via calling the LIB_OBSORT subroutine. The latter form is normally used, and is found in the OBSPROC (MAKECMA and FEEDBACK) and IFS/Screening. We have five different modes of the OBSORT:

• Mode 0, submodule BUFRsort. Partitioning and splitting of the BUFR data among the available processors (no geographical order, though).

• Mode 1, submodule CMA+BUFRsort. Geographical re-ordering of the CMA data in conjunction with the counterpart BUFR data; it is consequently called the CMA-data-driven BUFR sort.

• Mode 2, submodule CMAsort. Geographical re-ordering of the CMA data.; it also copes with the virtual-processor case where fewer processors than are required by the main analysis can produce more CMA files than the actual number of processors used by the OBSORT.

• Mode 3, submodule MATCHUP. Matching up and updating the ECMA data present in one geographical distribution with the CCMA data present in another distribution.

• Mode 4, submodule VMATCHUP. The same as MATCHUP, but for virtual processors. More than NPROC ECMA files are brought back to the NPROC files in the same order as the BUFR counterparts were left after the MAKECMA.

Next Section
Previous Section

20.03.2002

DATA ASSIMILATION

IFS documentation Front Page

Table of contents

9.1 Introduction