* Simple ring architecture * Replication using peer-to-peer protocol rather than leader selection * Linear scalability (99%) * Configurable consistency
How has it helped my organization?
Ability to achieve write speeds 10k tps: Compared to existing, it is 300% percent higher.
What needs improvement?
Row-level locking is not available; might be very helpful in update use cases.
Which solution did I use previously and why did I switch?
The earlier solution used Couchbase, which has leader selection. At times, when leader selection takes time, then we would lose the transactions. This got resolved with the peer-to-peer architecture solution in Cassandra.