VM Rightsizing Advisor
Forecast each VM's peak CPU, recommend a smaller size, then replay history to count how often that would have throttled.
Savings vs throttling risk each dot is a policy, backtested on 3 rolling origins
x: share of VMs whose throttled intervals exceed the SLO (0.5% of 5-min intervals). y: share of provisioned cores removed. Up and left is better. Teal = forecast policies, amber = baselines that size from a single history statistic.
Real Azure trace vmtable (2.7M VMs) + one 3.75-hour readings shard
Loading…