Optimising cloud costs for growing companies
An approach to reducing cloud costs without losing performance: first insight, then clean-up, then structural adjustment.
You optimise cloud costs by first measuring where the money goes, then tackling unused and oversized components, and finally aligning usage and contract. Do this periodically, because your usage changes as the company does.
Who is it for: Growing companies and teams that already use cloud infrastructure and want more control over the bill.
Where do you start with optimising?
This article is about cloud as infrastructure: servers, databases, storage and data traffic. For the basics of cost types and a calculation model, see the article on cloud costs for start-ups. Here the focus is on what you do when the company is already larger.
Start with insight. Without labels per project or team you only know the total. Add labels, group costs by purpose and determine which three items weigh the most. That is where attention pays off most.
Which measures deliver the most?
Costs usually do not fall through one big measure, but through a series of small ones. The following order often works well in practice.
- Clear out what is not being used: idle servers, loose disks, old copies and test environments.
- Adjust the size: compare actual use of compute and memory with what you have set up.
- Move data that is rarely read to cheaper storage classes, if the provider has them.
- Limit data traffic through caching and by keeping data closer to the user or the processing.
- Look at whether committed usage for stable workloads is more favourable than on-demand usage.
When is committed usage sensible?
Many providers offer a favourable condition if you commit to a certain capacity for a longer period. That suits a stable base load, not usage that fluctuates strongly. The downside is that you are tied in. Check what happens if your needs fall: can you change the contract, resell it or not?
How do you prevent costs from coming back?
Optimising is a habit, not a project. Plan a fixed moment at which you compare the bill with your expectation, give every environment an owner and give new environments an end date. Also watch the trade-off with quality: do not cut back on backups or monitoring to make the bill look better, because the damage from an outage weighs more than the saving.
Measures and their trade-offs
| Measure | Possible gain | Look out for |
|---|---|---|
| Clearing unused items | Lower bill straight away | First check that something is truly no longer needed |
| Right-sizing | Smaller servers at low usage | Test for peaks before you scale down |
| Storage classes | Cheaper storage for rarely read data | There may be costs for retrieval |
| Caching | Less data traffic and load | Outdated data if set up incorrectly |
| Committed usage | Lower rates at stable usage | You are tied to a period |
Who is this less suitable for?
- Small environments with low costs: the time optimisation takes may exceed the gain.
- Teams that do not yet have insight into costs: start with labelling before you intervene.
- Products where availability is the priority: do not scale down without a load test.
Frequently asked questions
How often should I review the costs?
A short monthly check and a more thorough review after major changes in usage or product is in practice a workable rhythm.
What is right-sizing?
Adjusting the size of servers and services to actual use, so you do not pay for capacity you do not use.
Can optimising harm performance?
Yes, if you scale down too far. So measure before and after the change and keep a way back ready.
