Link: Demystifying Database Systems, Part 3: Introduction to Consistency Levels

CAP

PNUTS vs CAP

• Link: Problems with CAP, and Yahoo's little known NoSQL system
• Paper: PNUTS: Yahoo!’s Hosted Data Serving Platform
• Lecture Notes

(YCSB was developed for PNUTS)

• Systems that sacrifice consistency tend to do so all the time, not just when there is a network partition.
• Yahoo's PNUTS:
  • Relaxes consistency to "timeline consistency": Replicas may not be consistent with each other, but updates are guaranteed to be applied in the same order to all replicas.
  • Gives up availability: If the master replica for a particular data item is unreachable, that item becomes unavailable for updates.
  • It gives up consistency not for availability, but to lower latency.
    • Messages over WAN are costly

Why this paper?

• Same goals as Spanner & memcache: Multi-region data replication -- facing 100ms network delays
• Better semantics than memcache
• Simpler than Spanner

Setup:

• Regions all around the world
• Each region has web servers & storage
• Big point: Low latency via user contacting nearest region
• Each region has all data