...

1. Supports interactive transactions.
2. Supports long running report-like read-only queries via RO transactions, including reading from backups. 
3. RO transactions do not affect RW transactions in any way.
4. Provides serializable isolation (and even more)
   1. Single key partition reads are strictly serializable
1. RW txns are strictly serializable
2. RO txnt are serializable
5. Txn duration is 2 * ConsensusTime (replicate + commit)
6. Lock held duration is 2 * ConsensusTime (replicate + commit) + 0.5 * RTT (unlock)
7. Txn size is bounded by (configurable) lock table size.
8. Recoverable cascadeless CC.
9. Only transaction data is serialized in persistent storage.

...

The high contention scenarios are not desirable for suggested locking based CC, but can be mitigated by using appropriate deadlock prevention strategy, like NO_WAIT_DIE.

In a nutshellfact, high contention scenarios are a sign of bad application design - they doesn't scale and render the clustering idea senseless.

We can reason more precisely on the performance as soon as some benchmark results are ready. Next step can be decided after analyzing their results, like implementing the other kinds of CC family - TO/OCC.

Discussion Links

// Links to discussions on thedevlist, if applicable.

...