Burstable Instance
also called T-series, B-series
An instance that provides a low baseline CPU allocation and accrues credits while idle, spendable for short periods of full performance.
Priced for workloads that are mostly idle with occasional activity — development environments, small internal services, low-traffic APIs. They are dramatically cheaper than a dedicated instance of the same nominal size, and for the right workload that is real money.
The failure mode is specific and produces confusing incidents. When credits are exhausted, the instance is throttled to its baseline — which may be 10% or 20% of a vCPU. The symptom is a service that performs normally for weeks and then becomes catastrophically slow, with CPU metrics showing low utilisation because the instance is not permitted to use more. Teams investigate the application for hours.
Two guards. Alert on credit balance, not just on CPU — a declining balance is the leading indicator and it is not on any default dashboard. And understand the unlimited mode offered by some providers: it removes the throttle and charges for the overage, which quietly turns a cheap instance into an expensive one if the workload has outgrown the family.
The rule of thumb: burstable for genuinely spiky, low-average workloads; anything with sustained utilisation above the baseline belongs on a fixed-performance family, where the cost is at least predictable.