I have a 3 nodes RS on EC2. All the nodes are running on r3.2xlarge, which has EBS Optimized turned on.
Recently, AWS released the r4 series, which is described as a better version of r3 - faster memory, better networking, lower costs. However, the r4 series doesn't have the EBS Optimized option. It does have "12 Gbps of dedicated throughput to EBS".
Is it a good idea to move the nodes to the r4 series or the EBS Optimized option is crucial (and not overcome'd by the dedicated throughput to EBS)?
All current "4" instances including R4 are EBS optimized by default, as are the i3 (high I/O), p3 (GPU compute), and f1 (fpga).
It would be easier if AWS used the number to indicate release year, but it's just incremented when an instance type is improved.
A large MongoDB server could benefit from the fast SSDs and very high IOPS of the i3 instance store. An r4.2xl is $0.53c/h. The i3.xl is $0.31 and gets 2M IOPS, the i3.2xl is $0.62 and gets 4M IOPS. If you're I/O limited they could be provide significantly improved performance.
Unlike EBS which is persistent, the instance store can be lost if the instance fails, and is lost on stop/start - but not restart. That means you have to persist it some other way. Perhaps: