Jira issue "CASSANDRA-6487 <https://issues.apache.org/jira/browse/CASSANDRA-6487> Log WARN on large batch sizes" touches on many of the comments I've made. Unfortunately, it doesn't answer the questions.
Since I'm using prepared statements to send 10 - 700 byte rows or ~7K destined to one partition, I suspect the batch size calculation is broken. It would take exploding the data by 6 times before the log warning 8400 bytes. I'm only using single replication for the current tests. Troy
