SEEK

I have recently created several Kepler workflow that contain multiple
eml200datasources. In particular, see 'workflows/eco/IPCC_Base_Layers.xml' and
'workflows/eco/GDAL_h1K_NS_Composite.xml'. The first was created by seaching for
'IPCC' and dragging the 10 baseline historical (1961-1990) climate icons to the
graph pane and then attaching to ClimateFileProcessor actors to create *.asc
files for use in GARP. The second contains 12 Hydro1K datasources (6 for North
America and 6 for South America). These data sources are projected, rescaled,
and merged into 6 *.asc files with the same resolution as the the IPCC output files.

I have run into several problems with these workflows that are apparently due to
the multiple eml200 datasources. These problems include:

1) First of all, the workflows are somewhat awkward to create. One needs to
seperately drag each datasource from the search results to the graph frame and
then configure the actor. Configuration is somewhat confusing with a very large
collection of options. (At a minimum, we need documentation for all the choices
somewhere.) Maybe we need some sort of datasource array (or sequence) actor
which could represent multiple data sources?

2) When one drags a datasource to the graph pane, the actor turns red while
loading data/metadata from the ecogrid and then turns yellow when no longer
busy. The delay isn't too bad while building the workflow (and it is in its own
thread, so it doesn't stop all other activities). However, when you reopen a
workflow with multiple data sources, you have to wait for all the sources to
turn from red to yellow. This can take a ridiculously long time on the machine
that was not used to create the workflow. [The time may be very, very long the
first time because the data has not been cached locally. It took 4-6 HOURS to
open the Hydro1K workflow (which has very large files) on a new machine!] And
even though each of the sources has its own thread, when there are a number of
such actor, they grap so much CPU time that it is not practical to do anything
else on a PC.

If I remember correctly, the main reason for the activity when an eml2datasource
is dragged to the graph pane is to create the dymanic ports for all the columns
in a data table. But the datasources here are not tables and the output ports
are determined by the configuration settings. Maybe whatever is going on in the
threads when the workflow is opened should be deferred until the workflow is
executed? This would let a user open a workflow and examine and edit it without
waiting a very long time.

3) There seems to be some errors that occur when trying to open a workflow with
multiple data sources. The IPCC workflow will sometimes open OK, but more often
all the sources except one will eventually turn from red to yellow. One,
however, will remain red and the CPU usage will stay pegged at 100%. The
datasource that will not load is not always the same if you shut down java and
start again!

Dan Higgins