Distributed Data

Question 1

Read-after-write

Answer 1

“Read-after-write” is a strategy to get around eventual consistency in a distributed db.

Imagine you post something to a forum, you refresh the page, and don’t see it. This is eventual consistency happening.

Instead, when a client writes a record, the primary returns the record (and version #) in the response. When the client reads again, it passes this version #, and any secondary will only reply if it has that # or higher.

Question 2

Reasons for SQL vs NoSQL

Sample data for NoSQL

Answer 2

SQL:

• Structured data
• Strict schema
• Transactions
• Complex joins
• Clear patterns for scaling
• More established: developers, community, code, etc
• Index lookups are very fast

NoSQL:

• Semi-structured data
• Dynamic or flexible schema
• Non-relational data
• No need for complex joins
• Store many TB/PB of data
• Very data intensive workload
• Very high IOPS throughput

Sample data for NoSQL:

• Rapid ingest of clickstream or log data
• Leaderboard / scoring data
• Temporary data such as shopping cart
• Frequently accessed (“hot”) tables
• Metadata/lookup tables