tags: #distributed-systems #go #mit #mooc #scalability #public

• Challeneges
  • concurrency
  • partial failure
  • perfomance
• Infrastructure
  • storage
  • communication
  • computation
• Abstractions
• Implementations
  • RPC
  • threads
  • concurrency control
• perfomance
  • scalability
    • 2X computers -> 2X throughput
    • Better because programmers are expensive than computers
    • You can easily scake web servers. But scaling database is not that easy
• fault tolerence
  • availability
    • If something goes down there is something else
  • recoverability
    • Easy to recover after repair
      
  • Non volatile storage can be used.
    • But avoid if possible.
  • replication
    • Out of sync issues
  • If you have like 1000 computers you can have like 3 computers fail a day
    • Exact number not probably
  • Someone stepped on network cable
• consistency
  • strongly consistent
    • eg Fetch the most recent
  • weakly consistent
    • Don’t have super strong consisteny. But still there are some ideas on how to make this useful
• map reduce
• go

Backlinks