Taco bell programming

• Do not overthink things
• Use pre-existing basic tools
• Functionality is an asset but code is a liability

The task

• Write a webscraper that pulls down a million webpages
• Search all pages for references of a given phrase
• Parallelise the code to run on a 32 core machine

The solution

• EC2 elastic compute cloud?
• Hadoop nosql database?
• SQS ZeroMQ?
• Parallelising using openMPI or mapreduce?

The ingredients

• cat
• wget
• grep
• xargs

Download a million webpages

	cat webpages.txt | xargs wget
      

Search for references to a given phrase...

	grep -l reddit *.*
      

...and if found then do further processing

	grep -l reddit *.* | xargs -n1 ./process.sh
      

parallelise to run on 32 cores

	grep -l reddit *.* | xargs -n1 -P32 ./process.sh
      

Turning it into a program

	#!/bin/bash
	cat webpages.txt | xargs wget -P pages
	grep -l reddit pages/*.* | xargs -n1 -P32 ./process.sh
      

Version Control

But I don't need it!

• Because you in the past hates you in the now
• final2014revisedb2bbb.zip
• No, dropbox will not do
• Software: Subversion / GIT / Mercurial
• Websites: Sourceforge / Github / Bitbucket
• Your employer will be looking at your profile

Github

Look at it grow

Optimising your code

Profiling

• You cannot compile with your eyes
• Profiler is only way to understand your code
• Causes code to run veeeery slowly
• Use timestamps if you are lazy

Profilers

• matlab: profile()
• Java: yourkit
• Python: cProfile module
• C++: gprof

Profiling example

Testing

• You should be testing your code
• Eyeballing is a start
• Do not ignore misbehaving code
• Have a baseline to test against
• Unit tests
• Static code analysis

Pylint

Incentives

Parallelising your code

Problems fall into two categories:

• Pleasingly parallel
• Disconcertingly serial

Pleasingly parallel

• Rendering frames in computer animation
• Functional programming (map/reduce)
• Simulations for independent scenarios
• Genetic algorithms

Disconcertingly serial

• State based calculations
• Writing to a single resource (I/O bound problems)
• Single global weather simulation

Amdahl's law

the effort expended on achieving high parallel processing rates is wasted unless it is accompanied by achievements in sequential processing rates of very nearly the same magnitude.

Example: rendering an animation

• parallelise a single frame
• parallelise a sequence of frames
• parallelise the scenes