// 2. REPL_SYNC: ~50 puts/s (occasional errors - nested
// runtime exception)
This is the slowest, since we're running a 2-phase commit protocol across the cluster
// 3. REPL_ASYNC: ~50 puts/s (occasional errors -
// nested runtime exception)
You should get close to the number with LOCAL mode, e.g. ~ 5000
In both #2 and #3, it all depends whether you bundle updates by using transactions.
Without transactions #2 is going to kill you b/c we run a 2PC after each put().
With #3, if you don't use transactions, use at least message bundling by enabling the queue.