PDiffView:  A prototype for Viewing the Difference in Provenance of workflow runs
Pierrick Girard, University of Pennsylvania


Scientific workflow systems are becoming increasingly important for
managing in-silico experiments. Such experiments are typically
specified as directed flow graphs, in which the nodes are modules and
edges represent data flow between the modules. Each execution (a.k.a.
run) of an experiment may vary the parameters and data inputs to the
modules in the specification; furthermore, alternative paths of the
workflow may be followed.  Comparing workflow executions of the same
specification and understanding the difference between them both
structurally and in terms of parameters and inputs is thus of
paramount importance to scientists for understanding the provenance of
final results.

We present a prototype system called PDiffView which allows users to
view, store and import/export workflow specifications and their
associated runs, for specifications which can be captured as
series-parallel graphs. The user may then see the difference between
two runs of the same specification by stepping through the set of edit
operations in the minimum-cost edit script, or by seeing an overview.
Since the graphs can be large, users may successively cluster modules
in the specification to form a hierarchy of composite modules. The
difference between two runs of that specification can then be viewed
at any level in the defined hierarchy, giving the user the ability to
zoom in on composite modules that indicate a large amount of change
and ignore others that indicate no change.