Transcript Slide 1

GungHo!
A new dynamical core for the Unified Model
Nigel Wood, Dynamics Research, UK Met Office
© Crown copyright Met Office
Outline

Unified Model – where we are now & the
need for change

GungHo!

Some results from each workpackage

Summary
© Crown copyright Met Office
Current Unified Model
“New Dynamics”
Davies et al. (2005)
Dynamics:
Physics:
• Regular lat/lon grid.
• Spectral band radiation
• Non-hydrostatic dynamics with a
deep atmosphere.
• Diagnostic or prognostic cloud
• Semi-implicit time integration with
3D semi-Lagrangian advection.
• Mass flux convection
• Atmospheric tracer advection
• Mixed-phase ppn
• Boundary layer
• Gravity wave schemes
Coupling possible to non-atmospheric components:
• Land surface model
• Sea ice model
• Ocean model
• Chemistry/aerosol model
© Crown copyright Met Office
Relative performance
450
400
350
300
250
Met Office
ECMWF
USA
France
Germany
Japan
Canada
Australia
200
150
20
11
01
20
10
01
20
09
01
20
08
01
20
07
01
20
06
01
2003
20
05
01
20
04
01
20
03
01
100
2011
3 day Northern Hemisphere surface pressure errors
© Crown copyright Met Office
Scalability
(17km)
Perfect scaling
T24/TN
24 nodes
© Crown copyright Met Office
(1 node=32 processors)
Nodes
The finger of blame…
 At 25km
resolution,
grid spacing
near poles =
75m
 At 10km
reduces to
12m!
© Crown copyright Met Office
Challenges!
 Scalability – remove the poles!
© Crown copyright Met Office
Challenges!
 Scalability – remove the poles!
 Speed – cannot sacrifice this for low resolution
moderate core counts
 Accuracy – need to maintain standing of model
 Space weather  600km deep model…
 Danger:
© Crown copyright Met Office
Everything to everyone…or
Nothing to anyone?
GungHo!
Globally
Uniform
Next
Generation
Highly
Optimized
“Working together harmoniously”
© Crown copyright Met Office
5 Year Project
 “To research, design and develop a new dynamical
core suitable for operational, global and regional,
weather and climate simulation on massively parallel
computers of the size envisaged over the coming 20
years.”
 Split into two phases:
2 years “research” (2011-13)
3 years “development” (2013-2016)
© Crown copyright Met Office
UK Collaboration of GFD,
numerical and computational
scientists
 5 FTEs from Met Office
(Dynamics Research; HPC Optimisation; UM
infrastructure)
 5 FTEs from NERC
(Bath, Exeter, Imperial, Leeds, Manchester,
Reading, Warwick)
 2 FTEs from STFC (Hartree Centre)
© Crown copyright Met Office
GungHo Issues
 How to maintain accuracy of current model on
a GungHo grid?
 Principal points about current grid are:
 Orthogonality
 C-grid
 These provide a number of good numerical
properties (Staniforth & Thuburn QJ 2012)
 Challenge is to retain those on a nonorthogonal grid
© Crown copyright Met Office
Some workpackage results
© Crown copyright Met Office
C-grid dispersion relations
Frequency
Grids
 Good dispersion
Exact
Low order
FEM
FD
 Minimal grid imprinting
Higher order
FEM Partially mass
lumped FEM
 Finite element approach
 Focus on: Cubed-sphere;
possibly triangles
Group velocity
 No computational modes
Cotter (Imperial), Melvin & Staniforth (MetO)
© Crown copyright Met Office
Nondimensional wavenumber
Grids
Higher order FEM
 Good dispersion
 Minimal grid imprinting
 No computational modes
U
V
 Finite element approach
 Focus on: Cubed-sphere;
possibly triangles
Cotter (Imperial), Melvin & Staniforth (MetO)
© Crown copyright Met Office
Partially mass lumped FEM
Φ
Recent results
Thuburn (Exeter)
Williamson Test Case 5 with 160K d.o.f.s (320x160)
9 m
6 m
FEM Cubed-sphere
FEM Hexagonal
10 m
ENDGame lat-lon
© Crown copyright Met Office
11 m
ENDGame rotated lat-lon
Are implicit schemes viable?
Weak horizontal scaling for a 3D Helmholtz problem
 Baseline resolution =
64x64
 Nz=128
 Grid cells per
processor = 520K
Algebraic Multi-grid
 Cs*Dt/Dx=const=8.4
 One side of cubedsphere
Conjugate Gradient
Geometric Multi-grid
Mueller & Scheichl (Bath)
© Crown copyright Met Office
Hector
What to do if not…
 Horizontally Explicit
– Vertically Implicit
(HEVI)
 Computational
modes arise from
multistep schemes
 Examine range of
Runge-Kutta
Implicit-Explicit
(IMEX) schemes
Weller (Reading) & Lock (Leeds)
© Crown copyright Met Office
HEVI
Implicit
Test cases
 Finite difference
scheme applied on
a variety of grids
 Simple solid body
rotation (Williamson
test case 2)
 Height and velocity
errors after 5 days
 Weller, Thuburn and
Cotter, MWR, 2012
Weller (Reading), Thuburn (Exeter) &
Cotter (Imperial)
© Crown copyright Met Office
Computational Science
Ham (Imperial), Ford & Pickles (STFC), Riley (Manchester)
 Vertical
loop inner
most
 Indirect
addressing
for
horizontal
 F2003
© Crown copyright Met Office
Transport
 Mass conservation = #1 user
requirement!
 Inherent part of mimetic approach
 But want to maintain non-split approach
of current SL scheme
 OK in horizontal (CFL<1 on uniform
mesh) – see previous simulations
 Challenge is in vertical…
© Crown copyright Met Office
Timetable…
 Further development and testing of
horizontal [2013]
 Testing of proposals for code
architecture [2013]
 Vertical discretization [2013]
 3D prototype development [2014-2015]
 Operational…by 2020
 Long term step change in scalability
© Crown copyright Met Office
Thank you!
Questions?
© Crown copyright Met Office