Comparison
RightsizingvsVertical autoscaling
Rightsizing
you look at a fortnight of real usage and cut the requests in half, and nothing at all happens.
Matching provisioned resources to observed usage plus deliberate headroom. It is unglamorous, repeatable and usually the single largest available saving, and it decays continuously as workloads change, so it is a recurring exercise rather than a project. The risk is rightsizing to an average and discovering the peak, which is why the percentile you size to should be a decision rather than an accident.
Full entry →Vertical autoscaling
something watched the workload for a week and rewrote its requests, and traditionally had to restart it to apply them.
Adjusting a workload's requests and limits automatically from observed usage, rather than changing how many copies there are. It solves the request-guessing problem that wrecks bin packing, and it is most useful on workloads that cannot be scaled out at all. Historically it required a restart to take effect, which is why many teams run it in recommendation-only mode and let humans apply the numbers.
Full entry →