[Mitgcm-support] AIX/obcs/version questions

mitgcm-support at dev.mitgcm.org mitgcm-support at dev.mitgcm.org
Wed Jul 9 15:45:21 EDT 2003


Hi, I've got a multi-part question, although there may be a single
answer...
I'm encountering problems with some runs on a Boston Univ. parallel machine
(twister.bu.edu: IBM p690 running AIX 5).
The configuration has a dense inflow at the top of a N-S slope and a
radiation boundary condition in the west. Closed walls everywhere else. It
runs fine on a 4-processor SGI (ceto.whoi.edu) but on the IBM it does funny
stuff at the boundary (creates dense water where there should be none...at
first this is confined to the boundary cells only but problems occur later
when the dense plume gets to the edge and wants to leave). I've made a few
modifications to the OBC calculation code, but it's the same on both
machines. I've also tried using the same 16-subgrid decomposition on the
SGI and that still works. Using advection 77 on both machines.
Two unresolved issues:
1) I'm not entirely sure that the genmake script I'm using is really the
right thing for this machine. Basically I've just stuck in the AIX case
Alistair suggested on 5/8/03 and it  mostly seems to work except for this
OBC thing. Would it be worth fiddling with this some more?
2) I'm not exactly sure which version of the code I am using or which one I
should be using (or even whether it's likely to make a difference). I first
downloaded to the IBM using "cvs co -d mit -P MITgcm" on May 8. I think it
was release1_beta1 that has been working on the SGI but I can't get that to
run on the IBM (although it might just be the older genmake...) I did just
download release1_p8, which the web page suggests is the newest "official"
version. I've compiled it (on the IBM) and my initial test still appears to
give weird behavior at the open boundary.
ok, this is kind of a mess. Any hints about what to try next? Anything
obvious come to mind? Why might the same code work differently on the SGI
vs. IBM?

cheers,
james






More information about the MITgcm-support mailing list