What is Autoscaling?
Autoscaling, told as a story about calling in staff — or sending them home — automatically, based on how long the line is.
Guests suddenly flood in, but the staff count stays the same.Without more hands, the line just keeps growing.
Size staff for the average, and you're short when busy, idle when quiet.One fixed number can't fit both.
Watch the line length, and automatically call in more staff or send them home.More when it's busy, fewer when it's quiet.
The staff count follows the same shape as the guest count.More guests, more staff — fewer guests, fewer staff.
Scaling up takes time, so the gap in between is still rough.That's why some systems predict ahead and scale early.
Autoscaling = watching the queue (or a gauge) and automatically adding or removing staff (servers) — always within a set max and min.
A technique that automatically adds or removes servers to match traffic. It can mean scaling out (adding more servers) or scaling up (making a server bigger), decided by watching metrics. Flipping too often causes thrashing, so systems set a max/min and a cooldown period.
When grown-ups say it
- Autoscaling
- Automatically adding and removing staff. It decides by watching the queue or a gauge.
- Horizontal / Vertical scaling
- Two ways to add more hands. Hire more people (horizontal), or make one person work harder (vertical).
- Scaling metric
- The number that decides when to scale. Things like queue length or a gauge reading.
- Scale-out delay
- How long until a new hire can actually help. The gap in between is still rough.
- Thrashing
- Scaling up and down too often. A cooldown period keeps this from happening.
- Max / Min limits
- The fence that keeps scaling from going too far. Scaling only happens within these bounds.
- Predictive scaling
- Scaling up early, based on a prediction. This connects to estimating next year's guest count in advance.
- Orchestration
- The control room deciding how many to run. It's what actually carries out the autoscaling decision.