[
https://issues.apache.org/jira/browse/AIRAVATA-344?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Suresh Marru closed AIRAVATA-344.
---------------------------------
Resolution: Fixed
Current implementation is working as described in this issue. Instead of the
load balancer, workflow engine submits to a RabbitMQ queue and GFac instances
use the work stealing approach to process them.
> [GSoC] Load balancing Workflow execution
> ----------------------------------------
>
> Key: AIRAVATA-344
> URL: https://issues.apache.org/jira/browse/AIRAVATA-344
> Project: Airavata
> Issue Type: New Feature
> Reporter: Saminda Wijeratne
> Labels: gsoc2012, mentor
> Fix For: WISHLIST
>
>
> Workflows are handed over to the GFac servers in order to be scheduled to be
> executed. Currently the user have to specify which GFac server should execute
> this workflow and the GFac server has a primitive scheduling mechanism to
> schedule the worksflows and its tasks. Large number of job submissions would
> require intelligent job submission from users end for the executions to yield
> efficiently. With some effort we should be able to get Airavata itself to
> perform better scheduling tasks. This can be in several layers.
> GFac server as a cluster: User will submit the workflow execution to a
> cluster rather than a single GFac server. The cluster will decide which GFac
> instance is most suitable for the workflow to be scheduled based on some cost
> parameter.
> Workflow scheduling: The BE should estimate some cost of running a workflow &
> try to schedule the workflows in a way which will minimize the cost of
> running all the workflows currently in the queue.
> Executing services: The BE should determine which implementation of the
> service to execute for a workflow node given there are more than one
> candidate.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
