Make costs visible
- Labels on projects and resources, with an application label linked to your CMDB.
- Billing export to BigQuery, the source of truth for analysis and the dashboard. Google Cloud is also among the providers of the FOCUS format.
- Budgets and alerts at billing account and project level.
Pay less for compute
| Mechanism | Principle | When to use it |
|---|---|---|
| Sustained use discounts | Automatic discount on some machine families running a large part of the month | No action: factor it into calculations |
| Resource-based committed use discounts | Commitment to vCPUs, memory or GPUs in a region, for 1 or 3 years | Steady, predictable load |
| Flexible (spend-based) committed use discounts | Commitment to an hourly amount, applicable across projects and regions | Steady load that changes shape |
| Spot VMs | Spare capacity at a large discount, interruptible | Batch, CI/CD, interruption-tolerant non-prod |
As on Azure and AWS: clean up and rightsize first (Google Cloud Recommender suggestions are a good starting point), then commit.
BigQuery: the item that surprises
- In on-demand mode, each query is billed by data scanned: a
SELECT *on a large table can be expensive. - Partitioning and clustering tables so only useful data is read.
- A maximum bytes billed cap per query, to avoid accidents.
- For a heavy, regular load, compare with capacity mode (slots).
Storage and logs
- Cloud Storage: Standard, Nearline, Coldline and Archive classes, with lifecycle rules or Autoclass.
- Cloud Logging: exclusion filters on useless logs and right retention. Same logic as for Log Analytics.
- Orphaned persistent disks and snapshots to delete.
The same method, on every cloud
Understand, optimize, operate, stay in control: the approach is identical on Google Cloud, Azure and AWS. See also our Azure and AWS pages.