Minutes of the WF-WG, GGF12, 22.09.2004, 11:30 - 12pm
-----------------------------------------------------


Chair Jim Blythe


Minutes taken by Andreas Hoheisel (andreas.hoheisel@first.fraunhofer.de)


Presentations:
--------------


Jim Blythe (ISI): Introduction and resource allocation on workflows
- Presentation of a Pegasus workflow with 1202 nodes
- Workflow for resource allocation: Pro and contra about reasoning about
  the whole workflow for resource allocation.
- TBA vs. WBA algorithm for estimating workflow times

----------------

Carol Goble (University of Manchester, UK)
- Work done within the myGrid project

- Application: Williams-Beuren Syndrome (WBS), using data from web-based
  services, such as BLAST, InterProScan, etc.

- Technologies used: Scufl, Taverna, SOAPLAB, Freefluo

- Workflows in this domain have typically small input, but huge output,
  are very data intensive, and change in time ("ad hoc workflows")

- Workflow description: They started with WSFL but they gave up, because
  higher level view on workflows is not given! Now they use Scufl.

- In Taverna there you have access to approx. 300 domain services that
  mostly produce string output. Problems: Licensing, unreliability,
  description of services, ...

- For comparison: SRS wraps more than 1000 services

- Important issues: 
  - Shims: enable services to fit together
  - Provenance tracking
  - Interactivity

Questions:

- UDDI-Server? Registry within Taverna uses RDF descriptions.

- Other application of Taverna? Taverna has been used in a lot of other
  projects (e.g., Dilbert Cartoons Web Service:-)

Comment by Darren Pulsipher: Workflow descriptions must also be simple
for business workflows.

---------------

Darren Pulsipher (cadence): A Look at Industry Workflows

Business worfklows requirements:
- Highly repeatable, reliable, and measureable, 
- Heterogeneous resources, applications, and processes
- Conditional workflows
- Partially run or rerun parts of the workflow
- Logs and results for all stakeholders
- Focus on reduction in total cycle time (e.g., faster development of
  software, or enabling more software development cycles)


Typical business workflows:

- Software development workflow: Gather Code to Merge -> Build Software
  -> Test and Package Software

- EDA: Simulation -> validation -> physical verification -> file
  conversion (no Web Services, huge applications)

- Design flow: 300-400 steps, involves 50-100 tools, about 50% of the
  flow can be automated (reduction of costs about $5M if one day
  reduction of cycle time)

- IT Operations: Event driven workflows (resource alerts, custom
  events), back-office (backups, financial applications, database
  applications, etc...), re-provisioning (platform validation and
  imaging), data center workflows (moving, cleaning and validating data)


Questions?

- Steps of the workflows can be workflows by themselves. Optimizing
  relation master / sub workflow? Engine typically not aware of the sub
  workflows. But we are moving in this direction.


Discussion
----------


Jim Blythe:

Different languages in different industries -> we need also different
views on same workflows

Carol: IPR on workflows can be an important issue. 

Darren: We are selling workflows to customers. There is a marking place
for this.

Carol: Debugger of the workflow has a different view on a workflow than
users. Several levels of abstractions.

Jim Blythe:

Open (science) vs. closed (business) workflows?

Workflow, e.g., for buying a laptop in a company is specific to the
company you are working in, but the services you will probably use, such
as Dell and Apple, are open and not under your control.


Darren: In our examples, we don't care about services, workflow is
sequence of processes in the business world.

Carol: NeSC workshop last year about workflow, including a survey of
requirements for scientific workflows. Report should be made available
for the WF-RG.