CIS5930-02 High performance computing for scientific applications

Finals -- review

Important topics for the finals are given below.

MPI

(Around 40 % of the grades)

init/finalize/rank/size
send/recv/isend/irecv/waitall/wait
Communicators: duplicate, split
Topologies: Cartesian topologies and their use
Derived data types: vectors, structures

Linear algebra

(Around 30 % of the grades)

Matrix distributions
- Striped
- Checkerboard
- Block
- Cyclic
Matrix vector multiplication
Matrix multiplication
Caching: Effect of loop order, cache-aware, cache-oblivious

Domain decomposition

(Around 15 % of the grades)

Model: graph partitioning, edge-cut metric and its drawbacks, recursive bisection
Geometric methods
- CND
- Inertial
- Space-filling curves
Combinatorial
- LND.
- Kernighan-Lin/Fiduccia-Matheyses
Spectral
- Laplacian
- Fiedler vector
- Partitioning based on the Fiedler vector
Multilevel methods
- Three phases: coarsening,initial partitioning, multilevel refinement

Monte Carlo

(Around 15 % of the grades)

Monte Carlo integration
Traditional parallelization through replication
Quasi Monte Carlo, and difference in error rate between MC and QMC
Random number generation: Seed, cycle, period
Parallel random number generation
- Cycle division: Leapfrog, sequence splitting
- Parameterization: cycle (or seed) parameterization, iteration function parameterization

Last modified: 28 Apr 2003