T Instance Credit System

T Instance Credit System:

1. CPU Credits

• Each T instance earns CPU credits continuously at a fixed rate depending on the instance size

• These credits accumulate when CPU usage is below the baseline performance

• Credits can be spent when CPU usage exceeds the baseline

• Maximum credit balance varies by instance size

• Credits expire after 24 hours for T2, but never expire for T3 instances while running

1. Baseline Performance

• Each T instance size has a baseline CPU performance level

• For example:

  • t3.micro: 10% baseline CPU performance, earns 12 CPU credits per hour

  • t3.small: 20% baseline CPU performance, earns 24 CPU credits per hour

  • t3.medium: 20% baseline CPU performance, earns 24 CPU credits per hour

• Operating at or below baseline allows credit accumulation

• Operating above baseline consumes credits

1. Credit Earning Rate Examples:

• t3.micro: 12 credits per hour

• t3.small: 24 credits per hour

• t3.medium: 24 credits per hour

• One CPU credit equals one vCPU running at 100% utilization for one minute

1. Burst Behavior

• When workload requires more CPU than baseline:

  • Instance spends accrued credits to burst

  • Can burst up to 100% CPU utilization if enough credits available

  • Performance gradually returns to baseline when credits are depleted

• No performance degradation during burst-to-baseline transition

1. Launch Credits

• New T instances receive launch credits

• Helps provide good initial performance for new instances

• Launch credits are used first, before earned credits

• Launch credits that aren't used expire after 24 hours

1. Unlimited Mode vs Standard Mode

• Standard Mode:

  • Instance can only burst while it has credits

  • Returns to baseline when credits are exhausted

  • No additional charges

• Unlimited Mode:

  • Can burst beyond earned credits

  • Incurs additional charges for sustained high CPU usage

  • Useful for workloads requiring consistent high performance

  • Can be enabled/disabled at any time

1. Monitoring Credits

• CloudWatch metrics available:

  • CPUCreditUsage

  • CPUCreditBalance

  • CPUSurplusCreditBalance (Unlimited mode)

  • CPUSurplusCreditsCharged (Unlimited mode)

1. Credit Behavior During Stop/Start

• Standard Mode:

  • Accumulated credits are lost when instance is stopped

  • New instance starts with zero earned credits

  • Receives new launch credits

• Unlimited Mode:

  • Surplus credits are billed when instance is stopped

  • New instance starts with zero earned credits

  • Receives new launch credits

Best Practices:

1. Workload Matching

• Use T instances for variable workloads with occasional spikes

• Monitor credit balance to ensure adequate performance

• Switch to non-burstable instances if consistently high CPU needed

1. Mode Selection

• Use Standard mode for:

  • Predictable workloads

  • Development/test environments

  • Non-critical applications

• Use Unlimited mode for:

  • Applications requiring consistent performance

  • Workloads with unexpected spikes

  • When occasional additional costs are acceptable

1. Cost Management

• Monitor credit usage patterns

• Set up CloudWatch alarms for credit balance

• Consider switching to M instances if consistently using all credits