Scylla is making an effort to be compatible with Cassandra, down to the level of limitations of the implementation. How is it different?
Scylla most commonly uses fewer rounds than Cassandra to complete a lightweight transaction. While Cassandra issues a separate read query to fetch the old record, scylla piggybacks the read result on the response to the prepare round.
Scylla will automatically use synchronous commit log write mode for all lightweight transaction writes. Before a lightweight transaction completes, scylla will ensure that the data in it has hit the device. This is done in all commitlog_sync modes.
Conditional statements return a result set, and unlike Cassandra, Scylla result set metadata doesn’t change from execution to execution: Scylla always returns the old version of the row, regardless of whether the condition is true or not. This ensures conditional statements work well with prepared statements.
For batch statement, the returned result set contains an old row for every conditional statement in the batch, in statement order. Cassandra returns results in clustering key order.
Unlike Cassandra, Scylla uses per-core data partitioning, so the RPC that is done to perform a transaction talks directly to the right core on a peer replica, avoiding the concurrency overhead. This is, of course, true, if Scylla’s own shard-aware driver is used - otherwise we add an extra hop to the right core at the coordinator node.
Scylla does not store hints for lightweight transaction writes, since this is redundant as all such writes are already present in system.paxos table.
More on Lightweight Transactions (LWT)