chapter 1 cont'd

• systems of record aka source of truth is the main db
• derived data systems (more than what you initially thought it was)
  • cache
  • denormalized values
  • indexes
  • materialized view
  • models
  • dwhs
• hosting
  • selfhosting
    • -ve: if you have spiky activity in non-spiky time your machines sit idle
    • -ve: have to maintain the machines yourself
  • cloud -ve's
    • at the whim of cloud providers (cost, downtime, bug fixes, and customizations)
    • trust them to keep data secure
• cloud native: software built for the cloud service
  • +ve's:
    • better perf
    • quick scaling
    • big data support
• RDMA: remote direct memory access, 2 devices' memories talking to each other, optimized for low-latency and high-throughput
• cloud services add a layer of abstraction
• RAID: redundant array of independant disks
  • make disk storage more durable by making copies
• cloud systems separate storage from compute, while you can can have virtual disks set up you run into network issues and are better off using purpose built things
• distributed system: several machines communicating via a network
  • every node in a distributed system can and will fail at some point so will the network between each node
  • network calls are slower than calling a different process on the same machine
  • with modern hardware you can do more on single node
• SOA: service-oriented architecture, each node acts as both a service and client
• microservice: each service does one thing, gotta be small and fast
  • they add a ton of overhead, and complicate deployments, testing, and backward compatibility
  • technical solution a people problem - something to avoid for as long as possible or just never grow that big
• FaaS: function as a service
• HPC: high performance computing (supercomputing)
  • every x store current state to disk
  • if a node fails stop everything until fixed and start from last known state