Running a cluster

A BNDL cluster consists of one or more workers and a driver in a fully interconnected network. The driver node is where your ‘program’ lives, i.e. the node which puts the workers to work. Typically 1 worker per core is executed.

Driver nodes are (typically) accessable through a ComputeContext, (see Context). The basic steps to starting a driver, creating a compute context and running workers are described in Getting started.

The easiest way to do start a driver and create a compute context is to create a Compute shell through the bndl-compute-shell command which is available after installation of BNDL. Workers can be started with the bndl-compute-workers command.

Todo

Describe typicall setups, data locality, etc.