Cloud provider discount & payment / Google Cloud
Google Kubernetes Engine (GKE)
The most mature managed Kubernetes: Autopilot or Standard
GKE (Google Kubernetes Engine) is Google managed Kubernetes, built alongside one of the project’s earliest and deepest contributors. This is where the "GCP is strong on data and Kubernetes" reputation comes from: the managed control plane, node pool management, version upgrades, and autoscaling are all built in. We open the GCP project and act as payment intermediary, with no personal credit card and no identity documents. Once the account is live you create node pools and namespaces in the console.
Specs & pricing
| Mode | Node management | Billed on | Good for |
|---|---|---|---|
| Autopilot | Platform sizes nodes from pod specs | Pod CPU, memory, ephemeral storage | Elastic workloads, less ops |
| Standard (regional) | You manage node pools | Management fee plus node resources | Production with fixed machine needs |
| Enterprise | Regional clusters, fuller security policy | Management fee plus node resources | Organisations with compliance audits |
| Node machine types | e2 / n2 / c2d and others | Varies by machine and disk type | Start ordinary workloads on e2 |
What GKE is: how much is actually managed
The difference between managed and self-built is not the UI — it is who owns the failure domains. In GKE, Google runs the control plane: API server, scheduler, and state. You own the workload: images, replica counts, requests and limits, quotas. Nodes can be configured by Autopilot from pod specs, or kept in node pools you manage.
Three modes: Autopilot, Standard, Enterprise
| Dimension | Autopilot | Standard |
|---|---|---|
| Node management | Platform scales from pod specs | You own node pools and machine types |
| Billing | Pod CPU, memory, ephemeral storage | Management fee plus node CPU, memory, disk |
| Granularity | Node-level operations not exposed | Machine type, disk, and labels are yours |
| Good for | Elastic workloads, less ops | Production with firm machine requirements |
How GKE is billed: management fee, nodes, discounts
- Cluster management fee: per cluster per hour whether or not you run anything; regional clusters cost slightly more
- Node resources: CPU, memory, disk, and traffic actually in use, on Compute Engine logic
- Autopilot bills per pod: scaling to zero mostly stops paying for idle capacity, which suits frequent scaling
- SUD and CUD apply to nodes too: they are ordinary compute instances, so do not leave that discount unclaimed
Versions and release channels
- Channels differ in upgrade frequency, not in production readiness — both run real workloads
- Upgrades roll at the node pool level, so old pods keep serving during the roll
- Migrate deprecated APIs before jumping several major versions, or you will stall on an intermediate release
Regions and topology: regional clusters and APAC placement
| Region | Location | Good for | Notes |
|---|---|---|---|
| asia-east2 | Hong Kong, China | Online services for South China, Hong Kong, Macau | Use regional clusters in production |
| asia-east1 | Taiwan | Taiwan users, or a second APAC footprint | A cluster in each plus global LB spreads risk |
| asia-northeast1 | Tokyo, Japan | Japan and Northeast Asia | Nodes follow the cluster region |
| asia-southeast1 | Singapore | Southeast Asia | Cross-region scheduling needs separate mechanisms |
Side by side: GKE against EKS and ACK
| Dimension | GKE | EKS | Alibaba ACK |
|---|---|---|---|
| Least-effort mode | Autopilot, billed per pod | Fargate serverless shape | Managed node pools with elastic scaling |
| Version tracking | Fastest, deepest upstream role | Mature and conservative | Tracks upstream open source |
| Fits its own services | BigQuery, Vertex AI | Redshift, S3, SageMaker | MaxCompute, OSS |
How to open the account and run a cluster with us
- Describe the service and scale: workload type, replica count, and target region (Hong Kong, Taiwan, Tokyo, Singapore, or US West)
- We handle provisioning and payment: no personal credit card bound, no identity documents submitted
- Create a regional cluster first: nodes spread across zones, so control-plane availability comes for free
- Set requests and limits before enabling HPA — a pod without resource declarations cannot be scheduled correctly
Four common mistakes
- Running production on a zonal cluster: one zone failure affects the whole cluster, and there is little reason not to go regional
- No requests and limits: the scheduler guesses, HPA loses its basis, and node utilisation swings
- Never upgrading: fall too far behind and older node pools are phased out, turning into unscheduled downtime
- Forgetting the management fee is fixed: it is charged even at zero pods, so include it in any comparison
Typical workloads
Open a Google Cloud account with us
We handle official provisioning and payment verification — no personal card, no identity documents. Order directly in the vendor console once live.
More Google Cloud products
Related reading
AWS Shield Advanced: what is free, what is billed
Shield Advanced covering your WAF costs is true, with four boundaries: only protected resources, only standard features, and subject to a request ceiling. Bot Control, CAPTCHA and web ACLs above 1,500 WCUs are still billed separately.
Cloud CostAWS private pricing (EDP): threshold, risk, shortcut
Private pricing is the highest-threshold layer of the AWS discount system: a cross-service discount rather than a single-service commitment. Two traps come first - a shortfall payment if you miss the commitment, and mandatory enterprise support. And AWS states plainly that buying through a partner qualifies too.
Cloud CostMulti-cloud cost visibility: unify categories first
Each provider runs its own billing system, so there is no ready-made unified view. Most people reach for a tool first, which reverses the order: unify the cost categories and the tagging scheme first. Four accounting mismatches will make the numbers disagree.