Related Work

Writing the “related work” for a project called “distributed”, is a Sisypheantask. We’ll list a few notable projects that you’ve probably already heard ofdown below.

You may also find the dask comparison with spark of interest.

Big Data World

• The venerable Hadoop provides batch processing with the MapReduceprogramming paradigm. Python users typically use Hadoop Streaming orMRJob.
• Spark builds on top of HDFS systems with a nicer API and in-memoryprocessing. Python users typically use PySpark.
• Storm provides streaming computation. Python users typically usestreamparse.

This is a woefully inadequate representation of the excellent work blossomingin this space. A variety of projects have come into this space and rival orcomplement the projects above. Still, most “Big Data” processing hype probablycenters around the three projects above, or their derivatives.

Python Projects

There are dozens of Python projects for distributed computing. Here we list afew of the more prominent projects that we see in active use today.

Task scheduling

• Celery: An asynchronous task scheduler, focusing on real-time processing.
• Luigi: A bulk big-data/batch task scheduler, with hooks to a variety ofinteresting data sources.

Ad hoc computation

• IPython Parallel: Allows for stateful remote control of several runningipython sessions.
• Scoop: Implements the concurrent.futures API on distributed workers.Notably allows tasks to spawn more tasks.

Direct Communication

• MPI4Py: Wraps the Message Passing Interface popular in high performancecomputing.
• PyZMQ: Wraps ZeroMQ, the high-performance asynchronous messaging library.

Venerable

There are a couple of older projects that often get mentioned

• Dispy: Embarrassingly parallel function evaluation
• Pyro: Remote objects / RPC

Relationship

In relation to these projects distributed…

• Supports data-local computation like Hadoop and Spark
• Uses a task graph with data dependencies abstraction like Luigi
• In support of ad-hoc applications, like IPython Parallel and Scoop

In depth comparison to particular projects

IPython Parallel

Short Description

IPython Parallel is a distributed computing framework from the IPythonproject. It uses a centralized hub to farm out jobs to several ipengineprocesses running on remote workers. It communicates over ZeroMQ sockets andcentralizes communication through the central hub.

IPython parallel has been around for a while and, while not particularly fancy,is quite stable and robust.

IPython Parallel offers parallel map and remote apply functions thatroute computations to remote workers

>>> view = Client(...)[:]
>>> results = view.map(func, sequence)
>>> result = view.apply(func, *args, **kwargs)
>>> future = view.apply_async(func, *args, **kwargs)

It also provides direct execution of code in the remote process and collectionof data from the remote namespace.

>>> view.execute('x = 1 + 2')
>>> view['x']
[3, 3, 3, 3, 3, 3]

Brief Comparison

Distributed and IPython Parallel are similar in that they provide map andapply/submit abstractions over distributed worker processes running Python.Both manage the remote namespaces of those worker processes.

They are dissimilar in terms of their maturity, how worker nodes communicate toeach other, and in the complexity of algorithms that they enable.

Distributed Advantages

The primary advantages of distributed over IPython Parallel include

• Peer-to-peer communication between workers
• Dynamic task schedulingDistributed workers share data in a peer-to-peer fashion, without having tosend intermediate results through a central bottleneck. This allowsdistributed to be more effective for more complex algorithms and to managelarger datasets in a more natural manner. IPython parallel does not provide amechanism for workers to communicate with each other, except by using thecentral node as an intermediary for data transfer or by relying on some othermedium, like a shared file system. Data transfer through the central node caneasily become a bottleneck and so IPython parallel has been mostly helpful inembarrassingly parallel work (the bulk of applications) but has not been usedextensively for more sophisticated algorithms that require non-trivialcommunication patterns.

The distributed client includes a dynamic task scheduler capable of managingdeep data dependencies between tasks. The IPython parallel docs include arecipe for executing task graphs with data dependencies. This same idea iscore to all of distributed, which uses a dynamic task scheduler for alloperations. Notably, distributed.Future objects can be used withinsubmit/map/get calls before they have completed.

>>> x = client.submit(f, 1)  # returns a future
>>> y = client.submit(f, 2)  # returns a future
>>> z = client.submit(add, x, y)  # consumes futures

The ability to use futures cheaply within submit and map methodsenables the construction of very sophisticated data pipelines with simple code.Additionally, distributed can serve as a full dask task scheduler, enablingsupport for distributed arrays, dataframes, machine learning pipelines, and anyother application build on dask graphs. The dynamic task schedulers withindistributed are adapted from the dask task schedulers and so are fairlysophisticated/efficient.

IPython Parallel Advantages

IPython Parallel has the following advantages over distributed

• Maturity: IPython Parallel has been around for a while.
• Explicit control over the worker processes: IPython parallelallows you to execute arbitrary statements on the workers, allowing it toserve in system administration tasks.
• Deployment help: IPython Parallel has mechanisms built-in to aiddeployment on SGE, MPI, etc.. Distributed does not have any such sugar,though is fairly simple to set up by hand.
• Various other advantages: Over the years IPython parallel has accrued avariety of helpful features like IPython interaction magics, @paralleldecorators, etc..

concurrent.futures

The distributed.Client API is modeled after concurrent.futuresand PEP 3184. It has a few notable differences:

• distributed accepts Future objects withincalls to submit/map. When chaining computations, it is preferable tosubmit Future objects directly rather than wait on them before submission.
• The map() method returnsFuture objects, not concrete results.The map() method returns immediately.
• Despite sharing a similar API, distributed Futureobjects cannot always be substituted for concurrent.futures.Futureobjects, especially when using wait() or as_completed().
• Distributed generally does not support callbacks.

If you need full compatibility with the concurrent.futures.ExecutorAPI, use the object returned by theget_executor() method.