Skip to content

Latest commit

 

History

History
48 lines (46 loc) · 1.9 KB

22-7-10---dynamo.md

File metadata and controls

48 lines (46 loc) · 1.9 KB
Raw

Dynamo

Aug 10th 2022

  • A over C
  • optimistic replication technique
    • leads to conflict
  • Now conflicts need to be
    • Detected
    • Resolved
  • 2 questions ?
    • When to resolve them ?
      • whether conflicts should be resolved during reads or writes
      • usually reads are kept simple
      • BUT For amazon users -> where add / remove from cart -> top priority
      • hence, -> always writeable
      • hence -> complexity of conflict resolution to the reads
      • ensure that writes are never rejected.
    • Who resolves them ?
      • can be data store
        • can implement policies like LWW (Last write wins)
        • or application can do custom
  • Other principles :
    • Incremental scalability
    • Symmetry - all node have same responsibility
    • Decentralization - in favor of peer-to-peer techniques
    • Heterogeneity - work distribution must be proportional to the capabilities of the individual server
    • No focus on -> data integrity and security
    • Characterized as a zero-hop DHT
      • Each node maintains enough routing information locally to route a request to the appropriate node directly
      • DHT = distributed hash table
      • to avoid multiple hops
  • System Interface
    • get(key)
      • returns single or list of conflicting versions along with context
      • Dynamo applies - MD5 hash on key -> generates a 128-bit identifier
      • used to determine the storage nodes to serve from
    • put(key,context,value)
      • context -
        • encodes system metadata about the object
        • information such as the version of the object
  • Partitioning Algorithm
    • dynamically partition the data over the set of nodes
    • relies on consistent hashing
    • advantage -> departure/arrival of a node -> affects its immediate neighbors only & others remain unaffected.
  • Read later -> What is consistent hashing & how has Dynamo solved the challenges of basic consistent hashing ?
    • Virtual nodes