11

Capabilities:• Dynamic load balancing and static data partitioning

- Geometric, graph-based, hypergraph-based- Interfaces to ParMETIS, PT-Scotch, PaToH

• Graph coloring• Graph/matrix fill-reducing or locality-preserving ordering• iZoltan interface supports ITAPS mesh interfaces

Download via the Trilinos toolkit:• http://trilinos.sandia.gov

Further information: http://www.cs.sandia.gov/Zoltan• Karen Devine, kddevin@sandia.gov• Erik Boman, egboman@sandia.gov• Siva Rajamanickam, srajama@sandia.gov

Zoltan Combinatorial Scientific Computing Toolkit

22

Example: LCLS RF gun, colors indicate distribution to different CPUs. Fields are computed only in causal region, using p-refinement. (Courtesy SLAC National Accelerator Laboratory.)

Particle Partitioning

Field Partitioning

SLAC’s Pic3P accelerator simulation solves Maxwell’s equations • Field computation on fixed mesh• Particles moving through domain

Load balance using two different data decompositions• Fields partitioned with graph-based methods (ParMETIS)• Particles partitioned geometrically (Zoltan RCB 3D)

Enables scalable solution of larger problems: • 24k CPUs, 750M DOFs, 5B particles

Zoltan example: Partitioning for Particle-in-Cell methods

33

Zoltan’s Data-Structure Neutral Design Supports a Wide Range of Applications and Data Structures.

Multiphysics simulations Adaptive mesh refinement

Crash simulations

Particle methods

Parallel electronics networks

12

VsSOURCE_VOLTAGE

12

RsR

12 Cm012

C

12

Rg02R

12

Rg01R

12 C01

C

12 C02

C12

L2

INDUCTOR

12L1

INDUCTOR

12R1

R

12R2

R1

2

RlR

12

Rg1R

12

Rg2R

12 C2

C

12 C1

C

12 Cm12

C

Linear solvers & preconditioners