📂 RDBMS

• Schema-based structured tables
• Ex: mySQL, SQLServer, SQLite, Oracle
• Restrictions
  • Each row has a primary key that is unique
  • Needs foreign key to join on another table
• Problems: just not cut out for today’s workloads, which have
  • Large amounts of unstructured data
  • Lots of random reads and writes
  • Joins are not as frequent as other operations
  • Not built to scale out
    • Scale up: grow capacity of each cluster by replacing with more powerful machines → not practical because hits hardware limitations
    • Scale out: grow capacity of network by adding more clusters → needs a distributed key-value store!

📂 NoSQL

• Unstructured, no strict schema (some columns will have some missing rows)
• Joins not always supported

Often use column-oriented storage instead of row-oriented

• Helps index entries in column quicker
• Helps perform range queries much quicker (avoids fetching other irrelevant data in “row”)
• Maintains pointers across column entries to maintain “row”

🔥 Cassandra

• Distributed key-value store that uses consistent hashing for mapping key → server (partitioner)
  • Unlike Chord, however, each server has full membership information (no finger tables) because a data center has enough space for this → gives us O(1) network hops