The following describes the benchmark configuration, unless otherwise noted below.

Note that little difference was seen between db.m3.large and db.m3.xlarge for the benchmark. So this would indicate that it's not worth spending extra for db.m3.xlarge. It is unclear as to why this is so and why moving from db.m3.large to db.m3.2xlarge, which is 4 times the price yet gives 6 times the performance. If the instance types are compared, apart from doubling CPU and memory for each step, db.m3.2xlarge has High network performance. So it seems that this is the factor which allows the dramatic improvement in throughput for db.m3.2xlarge. Of course this comment is purely in the context of this benchmark. Another application may experience different behavior.