[kepler-users] Workflow with instances dependency

Luis Assunção lass at isel.ipl.pt
Wed Jun 13 15:01:33 PDT 2007



First of all thank you for your answer, but I have some questions about your

-Did you suggest that I develop new actors or there is some kind of actors
in Kepler, ready to use, with the functionality (task-assigner; Check-Task)
that you have proposed?

-On Check Task1 are you suggested I call a web service that only responds
when K-1 result is available?

-After I sent my previous message I have tried kepler to invoke Web services
but I found many limitations. First The Web Service actor only works with
simple data types and it don’t support asynchronous calls and callbacks, a
easy way to solve my problem: task3 will be notified on completion of


I appreciate If you can give me more some help 





-----Original Message-----
From: Ilkay Altintas [mailto:altintas at sdsc.edu] 
Sent: quarta-feira, 13 de Junho de 2007 1:28
To: Luis Assunção
Cc: Kepler-users at ecoinformatics.org
Subject: Re: [kepler-users] Workflow with instances dependency




You can have a workflow the has the instances of web services for item K of
your simulation under PN or SDF (given n below is known before scheduling)
and has monitoring actors (check and get results if necessary) for the tasks
in item K-1. So assuming that your Task 1 through 4 depends also on each
other and there are n items, one way to model it is:



                                                            Check Task1(K-1)
Check Task4 (K-1)



K-Assigner(0..K..n) -> Task1 (K) -> Task2 (K)-------------- -> Task3 (K) ->
Task4 (K) 


K-Assigner will keep assigning the new items as the item before is
processed. In this way, you can model it using web services as only web
services for one item will be used at one time. 


If you do this in PN, you might end up having thousands of web services at
the same time for Task 1, but you can get around it by limiting it in the
workflow by the K-Assigner actor.



There might be more efficient ways of doing the same thing by keeping two
instances running at the same time using PN. Something like below, but the
one below won't work.



K-Assigner(0..K..n) -> Task1 (K-1) -> Task2 (K-1) -> Task3 (K-1) -> Task4

                                  |                           |

                                  |                          V

                                  -> Task1 (K) -> Task2 (K)
------------------------> Task3 (K) -> Task4 (K)



Hope you get the illustrations correctly...




On Jun 6, 2007, at 5:44 PM, Luis Assunção wrote:



In my current research I have to modulate a simulation where I have
thousands of input items and for each item I have the following workflow.


Task1 ->Task2->Task3->Task4

But Task2 of item K depends of the completion of Task1 from item K-1 and
Task3 of item K depends of the completion of Task4 from item K-1


I am interested to use modeling based on Workflows and I am also interested
to support task execution as a Web Service.

I am tried to modeling using Kepler with Web Service actor but I think it
isn’t suitable to describe an workflow with thousand of instances (huge
graphical boxes) and worst I don’t know the best way to describe task
execution with dependencies from previous execution?


Can someone advise me about what is the best pattern to this case?


I appreciate in advance some advising.

Many thanks


Luis Assuncao

Email: lass at isel.ipl.pt




Kepler-users mailing list

Kepler-users at ecoinformatics.org





Lab Director, Scientific Workflow Automation Technologies (SWAT)

San Diego Supercomputer Center(SDSC), UCSD

9500 Gilman Drive, MC: 0505  La Jolla, CA  92093-0505

phone: (858) 822-5453                        fax: (858) 534-8303

web:  <http://users.sdsc.edu/%7Ealtintas> http://users.sdsc.edu/~altintas


-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mercury.nceas.ucsb.edu/ecoinformatics/pipermail/kepler-users/attachments/20070613/59f91a33/attachment.htm

More information about the Kepler-users mailing list