NCSA Perf Engr and Computational Analysis -- Scalability

Achievements in Scalability

Scalability can have different meanings and be measured in different ways. It may mean how well an underlying architecture can support parallel applications, or it may be taken to be a measure of the efficiency of the parallel algorithms and code themselves. The graphs presented here cover these categories from several perspectives.

SGI Origin Scaling for Density Functional Theory

2-25-00

Graphs show the scaling achievements using MPI processes. Source: NCSA staff.

ARPS on the 250MHz Origin2000: ARPS (Advanced Regional Prediction System) is an atmospheric model that uses adaptive gridding to represent multiple spatial scales. Timing results for NCSA's 128-processor Origin2000 are shown.
Scaling on the Origin2000: A variety of discipline codes: Several disciplines are represented in these graphs showing speed-up over one processor.
NT SuperCluster vs. Origin scaling: The CACTUS graphs show the scaling achievements using MPI processes on the NT SuperCluster and the SGI Origin 2000. Source: NCSA staff.
Performance analysis of RIEMANN code: The RIEMANN graphs compare both the per-processor performance and multiprocessor scaling (and total performance) of the SGI Origin2000 and the SGI/Cray T3E for the RIEMANN code. Source: NCSA staff.
O2K vs. T3E scaling comparisons 9-9-99: The ZEUS graphs show the scaling achievements using MPI processes. Source: NCSA staff.
Hybrid programming paradigm applied to a CFD computational kernel: The Hybrid Model CFD graphs demonstrate how distributed MPI processes can be integrated with the shared memory loop-parallel threads in a single application for purposes of allowing the programmer more control over the load balance of the algorithm. Source: NCSA staff.

Tuning Highlights | PECM | SCD | Scalability

Computational Analysis - Scalability

Achievements in Scalability