Print

[bd]

I'm trying to determine if I can use JPPF for some of my apps.  Any guidance would be appreciated.

Does JPPF have a mechanism for forcing a set of tasks to run on the same node? I.e. is there support for smp processing?

I didn't see anything like this in the documentation, but the DataProvider mechanism appears to partially address the issue I'm faced with. I often have applications with data objects that consume most of a node's memory. To fully utilize multicore nodes, I need to have several tasks sharing memory on the same node. I gather that DataProvider provides shared memory for multiple tasks, but doesn't appear to address the task location issue.


thanks,

bd

[lolo]

Hello BD,

JPPF doesn't have yet the means to force the node on which tasks are going to be executed. There is a feature request for this, and we are still working on it.
However, you can configure a node to run multiple tasks in parallel by setting the number of processing threads in the node's configuration file.
There are 2 way to configure it:
- statically, in the configuration file for the node, just set the value for the property "processing.threads"
- dynamically, you can change the number of parallel threads from the admin console: in the Node Date view, right click on a node, and you will have the option to set the number of threads.

This is very efficient, I am testing wit a quad-core machine and having 4 or more threads makes a huge difference.

I hope this helps,
-Laurent

[bd]

Thanks.  I'll monitor the status of the feature request.

Yes, running multiple threads on a multicore machine can make an enormous difference with some apps.

The ability to do smp processing is becoming  important because of the relative cost of quad-core processors and memory: the price/core for quad-core processors is quite low, but the price of large amounts of memory is not.  E.g.  I have apps that routinely use data structures that use more than 3 gig of ram.  To fully utilize four cores in a non-smp configuration requires 16 gig of ram, which is prohibitively expensive.  In my case, I can run four tasks on the same data structure, so that with an smp configuration, the cores are well utilized.  (Obviously not all apps can be parallelized this way.)  Of course using a non-smp configuration is much simpler, but given cost advantage, I've decided that it's worth the trouble.