| General Questions NCAR IBM Related Questions Terrain Regrid Datagrid Little_R Rawins MM5 Graph Conventions used in MM5 Trouble shooting and clues on various computers |
Frequently Asked QuestionsMM5
The fortran utility program, mm5-time.f can give a rough estimate for cpu time required to run an MM5 job with specified domain configuration on NCAR's Cray J90 using f90. However, if more accurate information is needed, you need to run MM5 one time step to get the complexity factor (CF) used in mm5-time.f. CF = (one time step CPU) / (number of
grid points) / (3**nestlevel) where the number of grid points is IX*JX*KX. For coarse domain, the nestlevel is 0. Alternatively, one can make use of the following numbers for NCAR's Crays: Cray YMP single processor
5.25 x 10 E-5 sec using cf77 (non-hydrostatic, with a cumulus scheme and simple explicit scheme
[IMPHYS=4 in V2]). These numbers represent CPU seconds required
to compute at a single grid point and for a single time step.
One can obtain an estimated CPU time required for a single model
integration and a specific model configuration by multiplying
the total number of grids (IX*JX*KX), and total number of time
steps [total forecast time/time step (TISTEP in the MM5's namelist)].
The fortran utility program, mm5-memory.f
can give a first-cut estimation for required memory. However,
if more accurate information is needed, you need to run MM5 one
time step to get the memory information. MM5 model is designed for real-data modeling studies and NWP. But people have successfully used the model for some idealized modeling work. Care must be taken in such an exercise. We have some guidelines for you if you would like to do so. Note that this is not a STANDARD use of the model, therefore support is limited, and you have to be responsible for making changes to the modeling system programs wherever is needed. We also suggest that you should first understand the modeling system as it is before you undertake such an exercise. Martin Leutbecker of Germany has provided us a tar file for undertaking
such an exercise. This tar file is available from NCAR's anonymous
ftp site, and the file name is ideal.tar. It works for V2
system. If you are interested in trying it in V3, try this
site. - Since version 3.7 (December 2004), MM5
can be run with 6 nests (so a mother domain and 5 nest levels).
If you want to run MM5 with more than 3-level of nesting, it is simple enough to modify the code to do so. First go to source code directory domain/drivers, and copy subroutine nstlev3.F to nstlev4.F. Then edit subroutine nstlev3.F, and find the line that calls nstlev4 and uncomment it. So now you should have a line of Fortran code that looks IF(NUMLV(IDLEV+1,L).GT.0)CALL NSTLEV4(IEXEC,L) Next, edit the new subroutine nstlev4.F. Change the first line of the subroutine to SUBROUTINE NSTLEV4(IEXEC,ICOARS) Also, you need to modify the Makefile in domain/drivers to include nstlev4.F. Now you have the code that can be compiled with MAXNES=5, and LEVIDN > 3. Similar changes can be made to compile and run MM5 code with more levels of nesting. If you running the MPP MM5 version of the code, you need to make one extra change (If you have previously compiled the code for MPP, do a "make uninstall" after the change below, and before you again do a "make mpp"): Change the following lines in MM5/MPP/mpp_objects_all: #ifdef NESTEDNEST_OBJ = initnest.o chknst.o \ nstlev1.o nstlev2.o nstlev3.o \ $(MP_STOTNDT) smt2.o bcast_size.o \ merge_size.o mp_feedbk.o \ rdter.o #endif to look like this: #ifdef NESTEDNEST_OBJ = initnest.o chknst.o \ nstlev1.o nstlev2.o nstlev3.o nstlev4.o \ $(MP_STOTNDT) smt2.o bcast_size.o \ merge_size.o mp_feedbk.o \ rdter.o #endif - 2-D arrays: Change NVARSX from the current value (which is 47 in V3.5) to the new number in include/parame. For example, if you want to add two 2-D arrays, set NVARSX = 49. Modify include/point2d.incl to define the new pointers. Note that where you add the pointer in common block /ADDR2 matters. If you add new ones, it is better to add to the end. A similar change needs to be added to point2dn.incl. - 3-D arrays: You can always add your own common blocks, but it is more involved. It is easier to add a pointer array to an existing common block. Example 1: POINTER (IAQCA,QCA(MIXM,MJXM,MKXM)),(IAQCB,QCB(MIXM,MJXM,MKXM)), 1 (IAQRA,QRA(MIXM,MJXM,MKXM)), (IAQRB,QRB(MIXM,MJXM,MKXM) So that the new lines look like REAL QCA,QCB,QRA,QRB,QNEWA,QNEWBPOINTER (IAQCA,QCA(MIXM,MJXM,MKXM)),(IAQCB,QCB(MIXM,MJXM,MKXM)), 1 (IAQRA,QRA(MIXM,MJXM,MKXM)), (IAQRB,QRB(MIXM,MJXM,MKXM)), 2 (IAQNEWA,QNEWA(MIXM,MJXM,MKXM),(IAQNEWB,QNEWB(MIXM,MJXM,MKXM)) Note that the new array is dimensioned by MIXM, MJXM and MKXM which are defined only if IMPHYS >= 3. Then add the pointer to the common block /ADDR1. Remember you'll have to add this new one in the order that the pointers are listed. In this example, it means right after pointer IAQRB, or COMMON/ADDR1/ IAUA, IAUB, IAVA, IAVB, IATA, IATB,IAQVA,1 IAQVB, IAQCA, IAQCB, IAQRA, IAQRB, IAQNEWA, IAQNEWB, IAQIA, 2 IAQIB, IAQNIA, IAQNIB, IAQGA, IAQGB, IAQNCA, IAQNCB Make a similar change in point3dn.incl. Example 2:
REAL QVA,QVB,TA,TB,UA,UB,VA,VB 1 IAQVB, IATNEW, IAQCA, IAQCB, IAQRA, IAQRB, IAQIA, IAQIB, 2 IAQNIA, IAQNIB, IAQGA, IAQGB, IAQNCA, IAQNCB Make a similar change in point3dn.incl. If you do this for both 2-D and 3-D pointer arrays, you won't need to modify anything else. Since the array is declared in the pointer include files (i.e., point3d.incl, etc.), it will be available in any subroutine that has (for instance): #include < point3d.incl > They can be passed from SOLVE, for example, through argument lists to subroutines that don't already have the above include line.
|