Information Technology Reference
In-Depth Information
Workflow definition
Global
parameters
time_step:10
Task a
J
a1
: (file_1,10)
J
a2
: (file_3,10)
J
a3
: (file_5,10)
Local
parameters
y:min=1, max=20,
step=2
Task
description
.
Input=file_$y
Input=$time_step
J
a10
: (file_19,10)
Task b
Local
parameters
z=1,8,20
J
b1
: (1,10)
J
b1
: (3,10)
J
b1
: (3,10)
Task
description
Input=$z
Input=$time_step
.
.
.
FIGURE 5.13
Illustration of workl ow parameters.
device and generate corresponding output data or a parameter sweep task
that generates multiple data based on various sets of parameters. These
outputs can be produced at different times. Some successor tasks may not
require to be processed until all these output data are generated. It can
process the output once it is available. However, ancestor tasks can pro-
cess the output data generated from a single parent differently, depending
on their requirements.
Three I/O models have been developed in the workl ow system to pro-
vide data-handling capabilities. These models are:
many-to-many
,
many-to-
one
, and
synchronization
. In Figure 5.13, there are two tasks: task
A
and task
B
.
They are connected by a data link. There are multiple subjobs in
A
and
each job produces an output. For the many-to-many model, task
B
starts
to process data and generates an output once there is an input available on
the data link. As shown in
Figure 5.14a
, four outputs generated by four
subjobs of
A
are processed individually by four subjobs of
B
. For the many-
to-one model, task
B
starts to process data once there is an input available;
however, the result is calculated based on the result generated by earlier
subjobs. As shown in Figure 5.14b, subjob
B
1 processes the output gener-
ated by sub-job
A
1. Once the output of
A
2 is available, sub-job
B
2 is cre-
ated and processes the output of
A
2 based on the output generated by
B
1.
For the synchronization model, task
B
does not start processing until all
the output is available on the data link. As shown in Figure 5.14c, there is
only one subjob in task
B
and it processes all outputs generated by subjobs
of
A
at one time.
Search WWH ::
Custom Search