Gentle users,
A bug was reported in the MPP restart mechanism having to do with the timing of restart output. A description of the problem, reported by Norm Henry at the Meteorological Service in New Zealand, is followed by the fix. The fix will be incorporated in the next release of MM5.
Rotang, Aug 20, 2001
> -----Original Message-----
> From: Norm Henry [mailto:Norm.Henry@met.co.nz]
> Sent: Wednesday, August 15, 2001 11:27 PM
> To: 'michalak@ucar.edu'
> Subject: Stop/restart problem with MPP
>
>
> Hi John,
>
> I've been having a problem with stop/restart MPP runs failing at the
> beginning of the restart. I found this with v3.3 and have just installed
> v3.4 and found the same thing. Attached are rsl.out files for
> both the stop
> and restart.
>
> The namelist file for this run specifies a stop at 1080 min. The coarse
> domain time step is 108 sec = 1.8 min. From what I see at the end of the
> rsl_stop file, the model seems to want to carry on for an extra timestep
> beyond 1080 - hence the XTIMR variable is at 1081.8. The restart file
> itself is correctly named -
>
> ./restrts/r-01-0001080-0000
>
> but MM5 fails on restart when it reads the file and finds that XTIMR is
> not what it expected. This seems to be specific to MPP as we haven't
> seen this problem for SMP stop/restarts.
>
> Any suggestions?!
>
> Thanks in advance...
>
> Norm Henry
> Meteorological Service of New Zealand
To: Norm Henry
Subject: RE: Stop/restart problem with MPP
Dear Norm,
It appears this is something that got updated in the non-MPP v3.4 code
last November but not in the corresponding MPP code. The non-MPP
version of domain/io/output.F was modified:
C OUTPUT.37
C-----OUTPUT FOR RESTART: OUTPUT.38
C OUTPUT.39
#ifndef MPP1 OUTPUT.40
IF((.NOT.IFSAVE).OR.(KTAU.LT.NINT(SAVTIM/DTMIN)))GOTO 80 <<<<<<< 07NOV00.366
DO 66 LLN=1,MAXNES OUTPUT.42
IF(IACTIV(LLN).EQ.1)THEN OUTPUT.43
C RESET FINE MESH XTIME TO COARSE (CORRECTS TRUNCATION DRIFT) OUTPUT.44
XTIME=XTIMC <<<<<<< OUTPUT.45
IUTSAV=51+(LLN-1) OUTPUT.46
But the MPP code still has what appears to be older code to determine
whether to write a restart (07NOV00.366) and it is also missing the
code to reset the fine mesh XTIME to the coarse XTIME (OUTPUT.45). The
following is in the file MPP/RSL/mpp_output_10.incl):
CSTART mpp_output_10.incl
IF((.NOT.IFSAVE).OR.(XTIME.LT.SAVTIM))GOTO 80 <<<<<<<<<
PRINT *,' TOTAL NUMBER OF DOMAINS POSSIBLE ON RESTART',
* ' OUTPUT IS ',NSTTOT
DO 66 LLN=1,MAXNES
IF(IACTIV(LLN).EQ.1)THEN <<<<<<<<<
IUTSAV=51+(LLN-1)
IF (SVLAST)THEN
WRITE(SAVENAME,1000)LLN,0,RSL_MYPROC
ELSE
WRITE(SAVENAME,1000)LLN,IFIX(XTIME+.001),RSL_MYPROC
ENDIF
1000 FORMAT('./restrts/r-',I2.2,'-',I7.7,'-',I4.4)
OPEN (IUTSAV,FILE=SAVENAME,FORM="UNFORMATTED",ERR=64)
CALL OUTSAV(IUTSAV,ALLARR(1,LLN),IRHUGE,INTALL(1,LLN),IIHUGE,
+ ALLFG(1,LLN),IFGHUG,INTFG(1,LLN),IFGIHUG,ALLFO(1,LLN),
+ IFOHUG,MAXNES,MKX,MIX,MJX)
PRINT *,' +++ OUTPUTTING DOMAIN ',LLN,
* ' OUTPUT FOR THIS DOMAIN WAS UNIT ',IUTSAV
CLOSE ( IUTSAV )
GOTO 65
64 CONTINUE
PRINT*,'ERROR OPENING ',SAVENAME,' FOR RESTART. CONTINUING.'
65 CONTINUE
ENDIF
[The fix is to modify the indicated lines in the mpp_output_10.incl file,
above, to match the lines in domain/io/output.F. -Rotang's note]
Thanks for reporting this.
John