Conversion of CDAS/Reanalysis from Cray to IBM SP/2 ---------- -- --------------- ---- ---- -- --- --- CDAS/Reanalysis Project started running on a CRAY Y/MP. Later, we moved to a CRAY J16. Now, our CRAY J's have been removed and have been replaced by an IBM SP/2. The conversion to the IBM was a major conversion as one might expect. The CDAS/Reanalysis conversion was tested by running the systems on both the Cray and IBM SP/2. For the final test, the systems were run for a month and the output were compared. After successful testing, we started running on the SP/2 version starting with the 5/1/2000 analysis. Changes All cos-blocked files will have f77-style blocking. In the binary files, the ibm-mainframe (S370) binary numbers will be replaced by single precision IEEE numbers. In the binary files, 64-bit binary cray integers will replaced by 4-byte integers. In the bufr files, there will be no deviation from climatology. (The climatology is not useful as it predates Reanalysis.) dgn.* (monthly mean 3d flux files) will have more variables. Optimal averaging will be discontinued. Grib files will have f77-style headers. (The headers speed up processing and are ignored by grib decoders.) SSI will run in single-processor mode (not multi-threaded). The SSI uses data differently depending on the order of the input, and the order is indeterminate with multi-threading, GRIB scaling factors of the covariances will be changed. The znl files will use f77 blocking and real*8 data. Updated optimal interpolation quality control table (should have been done earlier). Some Benefits of the Conversion The source code to the various post-processors (spectral->pgb, spectral->sigma grid, spectral->isentropic) have been made portable. The w3lib grib routines are also portable. Source code is or will be made available upon request. The reanalysis forecast model is more-or-less portable. The rotating archive will have more variables. CDAS will run when the Crays are decommissioned.