PiCloud

Amit Saha

http://echorand.me/2012/08/17/pyconau-2012-talk-on-picloud/

Intro

Commercial cloud service

“Worker process” oriented rather than hosting oriented

20 core hours per month included in free account

Python API “import cloud”

REST API for more general access

Transfers dependencies automatically

Pre-initialised with NumPy/SciPy

Can tailor requested compute resources to a particular workload

Can run in a mode that uses multiprocessing locally rather than the cloud

Example

cloud.call is the most basic API!

Also the candidates you would expect, like cloud.call

(problems with the video demo failing to display properly - sorted out by using a different player)

(My comment: you could probably map a concurrent.futures executor to this API pretty easily)

IPython notebook provided in repo for talk (see link above)

Other capabilities

Directed-acyclic graph of job dependencies (to pass data between jobs, map/reduce, etc)

Persistent data: cloud.files API to move data to/from the cloud and update it in jobs

More compex example: pyevolve

Shows automatic deployment of dependencies

Identifies and pushes Python files that are referenced locally

Retrieves CSV files created pyevolve

Publishing APIs via REST

Can designate a Python function to expose as a REST API

Environment

Automatic deployment only works for pure Python modules

Environments let you tailor what it is installed - non-Python tools - extension modules - Ubuntu-based

Management APIs

Query job status, list jobs, etc.

Resources

More links! (again, see link at top)

Q & A

Any other services like this? (Don’t know)

Security - SSH authentication

My Comments

Looks pretty interesting. Definitely aimed at the scientific crowd rather than the webhosting crowd, thus the different emphasis and “worker process” style API.

Comments powered by Disqus