Skip to content

Latest commit

 

History

History
91 lines (90 loc) · 1.66 KB

faking-it-en.md

File metadata and controls

91 lines (90 loc) · 1.66 KB

Easy parallelization ways

Problems

  • Marginal only improvement of command parallelization at the CPU level
  • Transistor heat
  • Stagnant CPU clock frequency
  • Multiple (unexploited) CPU cores

Parallelization is tough

  • Our conscious thought processes are sequential
  • The Amdahl law restricts performance speedup
  • The road is full of traps
  • Race condition
  • Deadlock
  • Livelock

The hard solutions

  • Multiple threads
  • OpenMP
  • Programming languages
  • Scala
  • Haskell
  • Clojure
  • F#
  • Erlang
  • R (parallel)

The easy solutions

  • Specialized platforms
  • Multiple processes
  • Multiple threads

Specialized platforms

  • Databases
  • Web-centric development
  • Cloud services

In the Unix context

  • Without dependencies
  • xargs -P
  • GNU Parallel
  • With dependencies
  • Pipes
  • make -j

GNU parallel examples

  • Parallelized image decompression
find . -name '*.jpeg' |
parallel 'djpeg -scale 1/16 {} |
  cjpeg -outfile thumb/{}'
  • Parallelized file analysis
cat access_log* |
parallel --pipe print_user_agent

Libraries map, filter, reduce

  • C++
  • QtConcurrent
  • Java 8
  • java.util.streams
sales.parallelStream().
	map(s -> s.price).
	average().
	getAsDouble();
  • .NET
  • PLINQ

Specialized libraries

  • AMD Framewave
  • Intel Integrated Performance Primitives
  • Cover:
  • Image and audio processing
  • Video coding and decoding
  • Signal processing
  • Cryptography
  • Compression
  • Image creation (rendering)
  • NAG: Numerical and statistical processing

Cloud services

Examples AWS:

  • Elastic map reduce (Hadoop)
  • Lambda (event-driven code)
  • Elastic transcoder
  • Kinesis (streaming big data processing)