[Mitgcm-support] RE: puzzle of the day.....

[mitgcm-support at dev.mitgcm.org]

martin seems reluctant to do a profile...

i just tried switching GM, IVDC, EOS and
get the following times for thermodynamics
(expected time is 3 secs based on
4 secs * 15/20 )

20s -----> 17s --------> 9s --------------> 9s
    -GM          -IVDC      EOS="linear"

so that helps, but it still looks 3 times too big.

more suggestions please!

chris
-----Original Message-----
From: Alistair Adcroft [mailto:adcroft at mit.edu]
Sent: Thursday, February 28, 2002 11:48 AM
To: Support at Mitgcm.Org
Cc: Martin Losch
Subject: Re: puzzle of the day.....


Can Martin to a profile for us?

A.

> martin is running the global90x40 test.
> 
> the thermodynamics block takes 20 secs for 30 time steps.
> 
> for exp2
> 
> the thermodynamics block takes 4 secs for 30 time steps.
> 
> the vertical grid is reduced from 20 levels to 15 levels
> in global90x40.
> 
> dynamics runs in 8 secs for exp2 and 6 secs for global90x40
> i.e. it scales roughly according to reduction in levels.
> 
> global90x40 uses GM, exp2 does not..., but turning 
> GM off ( useGMRedi=.FALSE. ) only reduces global90x40:thermodynamics
> time to 17 secs ( not 4! ).
> 
> the setups are otherwise very similar i believe
> 
> was ist los?
> 
> chris