Distribution strategies

• NOTE: PEAA has a lot of outdated ideas, and tech and techinques have changed

What

• Distributed processing is done by placing objects on different nodes(processes on different machines connected via networks)

• Benefits

  • load can be distributed between different nodes, better performance ie throughput

  • Easier to add nodes (horizontal scaling and load balancing needed)

  • Calls between nodes are transparent and easy to follow

• Dont distribute if you dont have to

  • This is not the case for a lot of cases, apart from fast performance requirements. As the benefits of distributing (see micros services) can out way the costs

  • Does apply to objects, this pattern has been deprecated (ie COBRA, DCOM etc). Instead use of messaging protocols, rest with json/xml, grpc are used now.

Types

• Local call

  • calls between components on the same node

  • very fast

• Remote call

  • calls between components on different machines

  • Calls need to be course grained

    • minimise remote function calls

  • slower

    • marshall and unmarshal data

    • data transfer over network (latency and unreliability/partitioning)

  • Now speed of calls of fast, so do a few does not matter, but if dealing lots of calls, despite the speed of the response, the law of big numbers will see slow down

• OOP