Parallel Flow in VisTrails

When dealing with large datasets, the ability to leverage multiple cores or machines allows to speed up workflow execution time. VisTrails provides support for parallelization of tasks using IPython.

Setting up an IPython cluster

This package uses the standard IPython.parallel machinery to execute jobs. You will need to create and configure the IPython profile that you want VisTrails to use.

VisTrails is capable of running the ipcontroller and ipengine commands for you to start a controller or a set of engines locally, but for more complex setups, you can run ipcluster yourself from a terminal with the necessary options.

Interacting with the cluster

In the Packages/Parallel Flow menu, you will find the following options:

Start new engine processes
Use ipengine to start multiple new workers on the machine, using the current IPython profile. These processes will be shutdown on exit, unless detached using Cleanup
Show information on the cluster
Indicates whether we are connected to an IPython controller, the number of engines in the cluster, and which processes were started locally. Lists some basic information on each engine.
Change profile
Selects a different IPython profile. Performs cleanup first if a cluster was connected.
Cleanup started processes
Disconnects from the cluster and “forgets” the processes that were started from VisTrails. It is either possible to terminate them, or to leave them running in the background (for example, to be reuse by the next VisTrails session).
Request cluster shutdown
Sends the shutdown signal to the controller, regardless of whether it was started locally or not. If accepted, the controller will ask every connected engine to terminate and exit.

Note that when VisTrails is exited, it will shutdown the engines that it started. If it started the controller, it will also be shutdown, along with every engine that might have connected to it from other machines. To prevent that, use the ‘cleanup’ button and choose not to stop them; they will detach from VisTrails and won’t be killed automatically. You will still be able to use the ‘cluster shutdown’ button explicitely.

The Map module

Map allows you to execute a single module or a Group in parallel using input values taken from a list. It works in exactly the same way as the regular Map module from the Control Flow package.

Contrary to the standard Map module, the elements of the list will be submitted to the IPython controller which will execute them in a load-balanced manner on the engines currently connected.