[Mitgcm-support] RE: first results glo6

mitgcm-support at dev.mitgcm.org mitgcm-support at dev.mitgcm.org
Wed Jul 9 15:45:52 EDT 2003


Hi Guys,

 Great stuff. The fields look beautiful! Any more output
 available? Any news on the float stuff?

 I took a quick look at

 http://www.ecco.ucsd.edu/internal/glo6/ecco_glo6_justification.html

 We need to be a bit careful/respectful in comparing against 
 previous efforts. I agree it is very important to do a run where 
 much more output is saved and made easily available for analysis. 
 That seems like a valid  point. But you might want to avoid getting 
 yourself/us into a mud-slinging match over "performance". In particular 

 The reality is
 1. The models scalability on the SP is OK but not outstanding.
    This will improve of course :-).
 2. The model timestep is a factor of 2-4 down on previous 
    calculations of this sort. So even with impressive scalability
    and good on-node flops the current setup is not perfect.
    This too will improve of course :-).

 I think a slightly better spin is to make it clear that taking
 ECCO + SDSC + Blue Horizon + MITgcm together we are at a point where it 
 is possible to do a number of different renditions of this high-res
calculation. 
 That for each rendition a high temporal resolution three-dimensional
state can be saved
 and made available for analysis, biogeochemical studies etc...
 Mick could probably provide some nice input here too.
 On the IBM/SDSC infrastructure side emphasizing the balanced nature of
a 
 tera-scale computational facility with a complementary long-time, large 
 storage and data distribution focus is a safer, less confrontational 
 and more defensible position.

 I know its a small point but it could be unfortunate
 if something that could provoke a negative reaction made it into
 the Big Blue PR machinery. Hard to control it after that!
  
 Anyhow its impressive to look at! Great work. 

 Chris




More information about the MITgcm-support mailing list