Optimization of Execution Parameters of Moldable Ultrasound Workflows Under Incomplete Performance Data

task graph scheduling, workflow, genetic algorithm, moldable
tasks, makespan estimation, performance scaling interpolation

Complex ultrasound workflows calculating the outcome of
ultrasound procedures such as neurostimulation, tumour ablation or photoacoustic imaging are composed of many computational tasks requiring
high performance computing or cloud facilities to be computed in a sensible time. Most of these tasks are written as moldable parallel programs
being able to run across various numbers of compute nodes. The number
of compute nodes assigned to particular tasks strongly affects the overall
execution and queuing times of the whole workflow (makespan) as well
as the total computational cost.
This paper employs a genetic algorithm searching for a good resource
distribution over the particular tasks, and a cluster simulator evaluating
the makespan and cost of the candidate execution schedules. Since the
exact execution time cannot be measured for every possible combination
of the task, input data size, and assigned resources, several interpolation
techniques are used to predict the task duration for a given amount of
compute resources. The best execution schedules are eventually submit-
ted to a real cluster with a PBS scheduler to validate the whole technique.
The experimental results confirm the proposed cluster simulator corresponds to a real PBS job scheduler with a sufficient fidelity. The investigation of the interpolation techniques showed that incomplete performance
data can be successfully completed by linear and quadratic interpolations
making a maximum mean error below 10%. Finally, the paper shows it is
possible to implement a user defined parameter which instructs the genetic algorithm to prefer either the makespan or cost, or find a suitable
trade-off.