On Jan 21, 2014, at 426PM, Greg Trasuk <tras...@stratuscom.com> wrote:
>> >> >> As a side question: Could something like the Hadoop YARN be implemented >> using River? >> Hi Jeremy, Yes, I think you should be able to do this. The "River" way I think is a bit simpler because it does not require the use of a Resource Manager or a Scheduler. The common pattern here is a worker pattern (check out an example here http://www.rio-project.org/examples/workflow/index.html) where Workers are hungry and take as much work as they can handle. Workers on faster machines naturally consume more, work can be transactional. Workers take work based on registration of a "template" (obeying class matching semantics), and are generally configurable based on the number of CPUs. A side suggestion: You could also use Rio (http://www.rio-project.org) for the development and deployment of your services. Rio provides an easy to use, out of the box approach to create a multi-module project that includes a testing framework all easily integrated with Maven. Rio uses River for dynamic discovery, remote events and leasing of resources through the network. HTH Dennis
