Neptune T3 Burstable Instance Class

In addition to fixed-performance instance classes such as R5 and R6, Amazon Neptune gives you the option of using a burstable-performance T3 instance. While you’re developing your graph application, you want your database to be fast and responsive, but you don't need to use it all the time. Neptune's db.t3.medium instance class is just what you should use in that situation, at significantly lower cost than the least expensive fixed-performance instance class.

A burstable instance runs at a baseline level of CPU performance until a workload needs more, and then bursts well above the baseline for as long as a workload requires. Its hourly price covers the bursts, provided that the average CPU utilization doesn't exceed the baseline over a 24-hour period. For most development and test situations, that translates to good performance at a low cost.

If you start with a T3 instance class, you can easily switch later to a fixed-performance instance class when you’re ready to go into production, using the AWS Management Console, AWS CLI, or one of the AWS SDKs.

T3 Bursting Is Governed by CPU Credits

A CPU credit represents the full utilization of one virtual CPU core (vCPU) for one minute. That can also translate into 50% utilization of a vCPU for two minutes, or 25% utilization of two vCPUs for two minutes, and so on.

A T3 instance accrues CPU credits when it's idle and uses them up when it's active, both measured at millisecond resolution. The db.t3.medium instance class has two vCPUs, each of which earns 12 CPU credits per hour when idle. This means that 20% utilization of each vCPU results in a zero CPU credit balance. The 12 CPU credits earned are spent by 20% utilization of the vCPU (since 20% of 60 minutes is also 12). This 20% utilization is thus the baseline utilization rate that produces neither a positive nor negative CPU-credit balance.

Idle time (CPU utilization below 20% of the total available) causes CPU credits to be stored in a credit balance bucket, up to the limit for a db.t3.medium instance class of 576 (the maximum number of CPU credits that could be accrued in 24 hours, namely 2 x 12 x 24). Over that limit, CPU credits are simply discarded.

When necessary, CPU utilization can burst to as high as 100% for as long as needed by a workload, even after the CPU credit balance falls below zero. If the instance sustains a negative balance continuously for 24 hours, it incurs an additional charge of $0.05 for every -60 CPU credits accrued over that period. For most development and test workloads, however, bursting is usually covered by idle time before or after the burst.

Using the AWS Management Console to Create a T3 Burstable Instance

In the AWS Management Console, you can create a primary DB cluster instance or a read-replica instance that uses the db.t3.medium instance class, or you can modify an existing instance to use the db.t3.medium instance class.

For example, to create a new DB cluster primary instance in the Neptune console:

Using the AWS CLI to Create a T3 Burstable Instance

You can also use the AWS CLI to do the same thing:

aws neptune create-db-cluster \
    --db-cluster-identifier (name for a new DB cluster) \
    --engine neptune \
    --engine-version "1.0.2.2"
    
aws neptune create-db-instance \
    --db-cluster-identifier (name of the new DB cluster) \
    --db-instance-identifier (name for the primary writer instance in the cluster) \
    --engine neptune \
    --db-instance-class db.t3.medium