Lecture 17: COPS, Causal Consistency

The lecture explores causal consistency in distributed systems, using the COPS system as a case study. It begins by outlining the goal of achieving local reads and writes in geographically distributed data centers, contrasting this with systems like Spanner and Facebook Memcached. Two straw man designs are presented: one favoring reads with eventual consistency and another using a sync operator to enforce order. The discussion covers version numbers, Lamport clocks, and conflict resolution strategies. COPS is then introduced, emphasizing client context and dependency tracking to maintain causal consistency without central logging. The lecture also addresses limitations, including potential cascading delays and the inability to account for causality outside the system's awareness.