Information Technology Reference
In-Depth Information
candidate paths satisfying the search criterion. Based on the scientist-side
context, for example, the DNA sequence at hand, the first candidate
workflow will be suggested in which the data format matches operation
169's input. The resulting routine actually is a concatenation of three
snippets, each of which is from a myExperiment workflow.
This example demonstrates that
S
0
allows us to discover global
relations spread in multiple workflows and originally not easy to
identify. Our experience working with caBIG community shows that
this feature is quite useful for scientists to explore best practices from
multiple colleagues and combine their experiment snippets into a more
comprehensive one.
Method for Relation-Aware and Cross-Workflow
Search
Here we describe the method for the relation-aware and cross-workflow
search. For simplicity Q2 is formulated as follows: given two operations
o
i
and
o
j
, a set of workflows
W
and the derived operation network
S
0
,
how to find a path in
S
0
that connects
o
i
and
o
j
and meets a certain
criterion, for example, it should cross the least number of workflows.
The crossing-the-least-workflow criterion is reasonable because
each time when two snippets from two workflows are to be concate-
nated, additional tuning such as data transformation and security
enforcement are needed. Therefore, a path crossing fewer workflows
is more desirable. Again, if we make this analogy that operations are
stops in a public transportation system and a workflow is a bus/subway
route connecting multiple stops, we prefer a path between two stops,
which crosses fewer routes, that is, with less transfer overhead.
Figure 8.15 is a summary of the approach proposed here. Three
matrices, that is, adjacent (A), reachability (R), and transfer (T)
matrices are calculated offline and sequentially with their definition
to be given later. When finding a path between two operations, from
matrix
T
, we know how many transfers are needed; by referring from
matrix
T
back to
R
and
A
, we obtain these paths. Here we describe the
method shown in Figure 8.15 step-by-step.
Step 1: Calculate an Adjacent Matrix.
W
¼f
W
1
;
W
2
; ...;
W
n
g
is the set of workflows we extract from
myExperiment.
