Performance

Replica Exchange

Generated via trestles-128-256-many-cycles.py

Generated via trestles-enmd-vs-repex.py

Generated via trestles-variable-pilots-variable-replicas.py

Simulation-Analysis

X-axis: x/y where x is the pilot size, y is the number of compute units
In this experiment, each compute unit performs one simulation. Hence, y is also the number of simulations.
Y-axis: Execution time (seconds)

Blue bars: simulation phase
Green bars: analysis phase

Each simulation done with 2 cores each. Analysis phase performed over all cores(=N).

X-axis: x/y/z where x is the pilot size, y is the number of compute units, z is the number of ensemble members 
In this experiment, the 'z' ensemble members are divided over the 'y' compute units for simulation.
y1-axis: Execution time for individual phases(seconds)
y2-axis: Total execution time for both phases

Blue bars: simulation phase
Green bars: analysis phase

Each simulation done with 1 core each. Analysis phase performed over all cores(=N).

The above graph shows the simulation-analysis pattern executing 1 .. 64 'idle' (0 sec simulation + 0 sec analysis) workloads on 1, 2, 4 and 8 cores.

The above graph shows the simulation-analysis pattern executing 1 .. 64 '10 second' (10 sec simulation + 10 sec analysis) workloads on 1, 2, 4 and 8 cores.

AllPairs

Generated via run_2015-05-19.py

Generated via run_042215.py

Generated via run2.py