Print

[dbwiddis]

I am trying to simultaneously execute two jobs.

I have Job A (2 tasks) and Job B (1 task), both set non-blocking. 
I have one driver and two nodes all on the same physical machine. The machine has 4 cores and can easily support CPU for both nodes.
I send both jobs to the driver.
Job A processes on one of the nodes, while the other node sits idle.  Polling the collector for job status returns EXECUTING for Job A and PENDING for Job B.
Only after Job A is complete does Job B begin processing on the same node.  The idle node never seems to be used.

On a second attempt with identical settings I got the same result (only one node processed both jobs, in series) but they used the other node, implying both nodes are properly configured to process the jobs/tasks, but the driver is only choosing one node to send both tasks to.

This is similar to the issue discussed in this thread, except that thread implies running two nodes will solve the issue, but it doesn't.  Is there another configuration parameter I'm missing?

Code: [Select]

15:32:25,641 INFO  [JPPFClient] - Job Status:  EXECUTING PENDING
15:33:25,642 INFO  [JPPFClient] - Job Status:  EXECUTING PENDING
15:34:25,642 INFO  [JPPFClient] - Job Status:  EXECUTING PENDING
15:35:25,643 INFO  [JPPFClient] - Job Status:  EXECUTING PENDING
15:36:25,644 INFO  [JPPFClient] - Job Status:  EXECUTING PENDING
15:37:25,645 INFO  [JPPFClient] - Job Status:  COMPLETE EXECUTING
15:38:25,648 INFO  [JPPFClient] - Job Status:  COMPLETE COMPLETE

[lolo]

Hello,

You also need to ensure that you have configured at least two connections to the driver in your client configuration. This is what will allow both jobs to be sent conccurently to the driver. Otherwise, the second job will remain in the client queue, while the other is executing. Could you please confirm whether you have done that in your client configuration file? Or, if you configured the connections programmatically, as in the thread you mentioned, could you confirm that it is done before the JPPFClient is created?

Thanks for your time,
-Laurent

[dbwiddis]

Laurent,

Thank you for this reply.  You were correct that I needed to specify a larger thread pool.   Setting pool.size correctly in my client properties file solved this problem.

I've made quite a bit of progress setting up my framework on the Rackspace cloud.  I would like to contribute a section to your documentation similar to the EC2 section you have.  Where should I post my lessons learned?

[lolo]

Hello,

Thank you very much for offering this contribution 

The most efficient way to proceed would be if you could register with our issue tracker and submit a feature request. You could provide your input as text or as an attachment in any format you're comfortable with. From there, I propose to integrate it in our documentation, most likely with a bit of reformatting but without changing the content, unless it makes obvious sense to change it. I can easily create a page on the web site that you will be able to review before we finalize it. Would this be ok for you?

Otherwise, you can simply post your input here in this thread, or send me a mail at laurent dot cohen at jppf dot org.

Sincerely,
-Laurent

[dbwiddis]

Thanks. Looking at the EC2 documentation there's a lot of overlap, so I propose to rewrite that section as "EC2 or Rackspace" and keep a lot of the existing, while adding some new information.   I'll add it to the issue tracker.

[lolo]

Hello,

I'm not sure if you received a notfication form the issue tracker for JPPF-258, so I'm posting here as well. I have integrated the new content into a new page, which you can read and review from here. I kept the old and new page side by side to make it easier to review. Would you mind providing feedback on the new page, if only to say it is ok to publish?

Also, would you like to be acknowledged for this contribution? For instance, I could add your name to the page or in any other place you see fit.

Thanks,
-Laurent

[dbwiddis]

Your link is from localhost, so I can't see it.  I guessed at a few URL's at port 8880 but they don't work...

[lolo]

Arrg, I'm so bad 
My apologies for this. Here is te correct URL: http://www.jppf.org/doc/v4/index.php?title=Deployment_and_run_modes

For the story, I'm keeping a replica of the JPPF web site (and the environment to run it) on a local machine, so I can test web site changes before publishing them. That's where the localhost url comes from.

-Laurent

[dbwiddis]

Looks good to me!  Go ahead and publish. I don't need any personal recognition!