Something

202401161252
Status: #idea
Tags: DC

Something

Simple Classification (Slide 3 TODO)

  1. High Performance Distributed Computing
    1. Cluster computing
      • Similar nodes, high speed network, same OS
      • Supercomputers
    2. Grid computing
      • Decentralized, plenty of nodes from everywhere, different admin domain
      • Open Grid Services Architecture (OSGA)
      • Fabric: Provides interfaces to local resources
      • Connectivity: Communication/transaction protocols
      • Eg. SETI
  2. Distributed information systems
    1. Distributed transaction processing
      • All or none of the requested are executed
      • Transaction Processing (TP) monitor -> Coordinates the execution of a transaction in different DB's
      • Eg. MakeMyTrip
    2. Enterprise Application Integration (EAI)
      • Letting applications talk to each other
      • Different communication methods
        • File transfer: Simple, update notification
        • Shared database: Much more flexible
        • Remote Procedure Call (RPC)
        • Message Oriented Middleware (MOM): Pub-sub model
  3. Pervasive computing
    1. Ubiquitous Systems
      • Eg. Autonomous vehicle (perfect example)
      • Everywhere, and always available
      • Interaction -> Highly unobtrusive
        • Eg. Smart, personalised car
      • Distribution -> Networked and distributed, accessible in a transparent manner
      • Context awareness
        • Eg. GPS-awareness
      • Autonomy
        • Highly self-managed
        • Automatic updates, IP address allocation, device drivers (when plugged in)
        • Eg. DHCP, Universal Plug and Play (UPnP)
      • Intelligence
        • Collects lot of data and uses ML models for real-time decision making
    2. Mobile computing systems
      • Devices change location over time
      • Stable communication is a big challenge
        • Possible solution: GMaps caches data offline
      • Directly sharing resources between mobile devices
        • Scrapped now, after lot of research. No luck
      • Pervasive, but inherently mobile
      • 2 types:
        1. Mobile cloud computing (TODO image from slide 10)
        2. Mobile edge computing (TODO image from slide 10)
          • Eg. Each telecom tower becomes and edge device
      • Bottom line -> Mobile devices connects to stationary servers
    3. Sensor networks
      • 2 views:
        • TODO Images from slide 11
      • Eg. What is the traffic load at a junction?
        • Distributed DB or centralised DB?
        • Sensing and aggregation can be local or centralised
        • Dealing with sensor node failures is challenging (in distributed model)
        • Lot of research to make smart sensors (sensor + compute). Leads into tinyML (not covered in DC) research.
      • Eg. TinyDB
        • Declarative database, with a single root for sensor data
        • Makes a tree structure
      • Alternative: Have special nodes t
      • INFO: Intel is researching a lot into drone computing.
      • INFO: Startup which works on sports analytics with drones and camera. Wants to collab with BITS Goa. Apparently best in India

The Cloud and The Edge

TODO Image from slide 12

Pitfalls of Distributed System Development

References