Rightsizing vs Burst
Bursty workloads.
Overview
Rightsizing matches instance capacity to workload pattern. Bursty workloads benefit from burst-credit instances; steady-state workloads need always-on capacity. The wrong choice either pays for unused capacity or starves the workload during peaks.
- Bursty workloads. Idle most of the time with occasional CPU peaks. T-class instances with burst credits fit the pattern.
- T-instance burst credits. AWS T3/T4 instances accumulate credits while idle and spend them during peaks. Cost-efficient for genuinely bursty load.
- Steady-state instances. M-class always-on. Workloads that sustain CPU pressure exhaust burst credits and slow down on T-class.
- Burst credit monitoring plus per-tier sizing. Track credit balance per T-instance; customer-facing tiers stay always-on, internal tiers can burst.
The approach
Three habits make capacity-class choice defensible: analyse the workload pattern, monitor burst credits as a standing signal, and review sizing quarterly.
- Workload pattern analysis. Pull 30 days of CPU utilisation. Bursty (idle plus peaks) selects T-class; steady-state selects M-class or larger.
- Monitor burst credits. CloudWatch CPU credit balance per T-instance. Alert when credits trend toward exhaustion.
- Per-tier sizing. Customer-facing instances always-on; internal batch jobs can burst. Match risk to capacity model.
- Quarterly review plus documented rationale. Walk the instance catalogue each quarter; per-instance rationale lives in the IaC commit.
Why this compounds
Each correctly-sized instance produces ongoing cost benefit. The team learns workload patterns through repeated review; new instances ship sized correctly from day one.
- Cost efficiency. Right capacity model matches workload. T-class for bursty, M-class for steady; both correctly sized.
- Performance. Right size avoids burst-credit exhaustion. The customer-facing tier never throttles unexpectedly.
- Operational fit. Per-tier sizing matches priority. Critical workloads get always-on; cost-sensitive ones get burst.
- Year-one investment, year-two habit. The first review is heavy lift. By year two the cadence runs itself and savings compound.