Adaptive Execution Planning in Biomedical Workflow Management Systems

Workflow management systems try to move grid,
cloud and high performance computing (HPC) services closer to
scientific and industrial community by providing a user-friendly
interface enabling definition of complex problems presented
as workflows. Workflows provide a formal way to define and
automate multi-step procedures reflecting real-world phenomena.
However, this still places demands on users to decide how to
execute particular tasks in workflows. k-Dispatch, a platform
providing automated tasks execution, planning and monitoring,
focuses on selected workflows from medical environment. For
security reasons, only in-house code binaries tuned for specific
HPC resources are used. k-Dispatch screens out users from
the complexity of HPC systems. This paper describes how the
presented framework deals with the task execution planning.
Static planning that uses default execution parameters may not
be sufficient for the effective execution since the results delivery
is time-constrained, and there is an effort to minimize the
computational cost. Adaptive planning discussed in this paper
may improve this process.