Card 09/ 09
Recap: Scaling Out
Up or out, what has to be stateless first, and what autoscaling cannot catch.
- A system scales when added resources buy proportional capacity; scaling up is a bigger machine, scaling out is more machines (Scalability: Handling More Load by Adding Resources).
- Scaling up changes no code and stops at the largest machine; each step about doubles the price (Vertical Scaling Example: A Bigger Database Machine).
- Scaling out needs stateless copies behind a load balancer, and has no fixed ceiling (Horizontal Scaling Example: Ten Stateless Application Servers).
- Stateless parts out, stateful parts up first, and the team's size counts (Scaling Up vs Scaling Out).
- Sessions in one server's memory log users out once there are several servers (Sessions in Server Memory).
- Autoscaling follows load by a rule, minutes behind it (Autoscaling: Capacity That Follows Traffic).
- A spike faster than a server starts needs spare capacity or scaling ahead (Autoscaling Too Slow for a Spike).
- System Design
- Performance
← → move through sections and cards