Introduction
Cloud bills grow quickly when resources are provisioned generously and left running indefinitely. Cost optimization is the ongoing practice of matching cloud spend to actual need, using specific, measurable levers rather than vague cost-cutting intentions.
Cricket analogy: A franchise that keeps every player on full salary year-round regardless of form sees its wage bill balloon, just as cloud bills grow when resources are provisioned generously and left running without review.
Explanation
Four concrete levers account for most cloud cost savings. Right-sizing means matching instance types and sizes to actual observed CPU, memory, and network usage instead of guessing—downsizing an oversized virtual machine that averages 10% CPU utilization is a direct, measurable saving. Reserved and spot instances trade flexibility for discount: reserved (or committed-use) instances offer a lower hourly rate in exchange for a 1- or 3-year usage commitment on steady-state workloads, while spot instances offer very deep discounts on spare capacity that the provider can reclaim with short notice, making them suitable for fault-tolerant or batch workloads rather than critical, always-on services. Storage tiering moves data to cheaper storage classes as it becomes less frequently accessed—for example, moving month-old logs from high-performance storage to a lower-cost archival tier—since paying premium storage rates for rarely accessed data is unnecessary spend. Auto-scaling avoids over-provisioning by automatically adding capacity when demand rises and removing it when demand falls, so a fleet sized for peak traffic doesn't sit fully provisioned—and fully billed—during quiet periods.
Cricket analogy: Swapping an oversized batting lineup that rarely uses its full depth for a leaner squad is right-sizing, locking in a season-long central contract is a reserved instance, using a low-cost domestic-league player for a fault-tolerant warm-up match is a spot instance, archiving old match footage to cheaper storage is tiering, and calling up extra fielders only during a big tournament is auto-scaling.
Example
resource "autoscaling_policy" "web_fleet" {
name = "web-fleet-scaling"
min_size = 2
max_size = 10
desired_size = 2
scale_up {
metric = "cpu_utilization_percent"
threshold = 70
adjustment = 2
}
scale_down {
metric = "cpu_utilization_percent"
threshold = 30
adjustment = -1
}
}Analysis
This auto-scaling policy keeps the web fleet at a minimum of 2 instances during quiet periods and scales up to 10 only when CPU utilization exceeds 70%, then scales back down once utilization drops below 30%. Compared to statically provisioning 10 instances around the clock to handle peak load, this configuration pays for extra capacity only while it is actually needed—directly reducing spend without reducing the fleet's ability to handle traffic spikes. Combined with right-sizing the base instance type, using reserved instances for the guaranteed minimum of 2, and tiering old logs to archival storage, these levers compound into substantial savings versus a static, generously-provisioned setup.
Cricket analogy: Keeping a minimum core of 2 net bowlers on hand and calling up to 10 extras only when a big tournament's workload exceeds 70% capacity, then releasing them once it eases, saves money versus contracting all 10 all season.
Key Takeaways
- Right-sizing matches instance types/sizes to actual observed usage rather than guesswork.
- Reserved/committed instances discount steady-state workloads; spot instances discount fault-tolerant, interruptible workloads.
- Storage tiering moves infrequently accessed data to cheaper storage classes.
- Auto-scaling avoids over-provisioning by matching capacity to real-time demand.
Practice what you learned
1. What does right-sizing a cloud instance involve?
2. Which workload type is best suited to spot instances?
3. What is the purpose of storage tiering as a cost lever?
4. How does auto-scaling help control cloud costs?
Was this page helpful?
You May Also Like
Auto-Scaling Basics
Learn the difference between horizontal and vertical auto-scaling and how scaling policies use metrics like CPU utilization to trigger scaling actions.
Cloud Storage Classes
Learn how hot, cool, and cold/archive storage tiers trade off cost against retrieval latency for different access patterns.
Virtual Machines in the Cloud
Learn how cloud virtual machines work, how to size them, and how on-demand, reserved, and spot pricing models trade off cost against reliability.
Cloud Computing Benefits and Challenges
Concrete benefits of cloud computing like elasticity and pay-as-you-go, weighed against real challenges like vendor lock-in and cost sprawl.