# GKE

Sedai continuously analyzes GKE usage and autonomously rightsizes pods and nodes across Standard and Autopilot, plus GPU and disk optimization.

## Right-Size Every GKE Workload, Autonomously

GKE removes the operational burden of running Kubernetes, but resource decisions like pod requests, node packing, and GPU allocation still drive your bill. Sedai continuously analyzes actual usage and safely adjusts resources in both Standard and Autopilot clusters, so you stop paying for capacity you don't use.

## GKE Makes Kubernetes Easier to Run. Not Easier to Right-Size.

Google manages the control plane, but the resource decisions that drive cost are still yours to get right, and most teams don't have the bandwidth to do it continuously.

### Workloads are over-provisioned by default.

CPU and memory requests get set from worst-case estimates and rarely revisited, wasting resources regardless of GKE mode.

### That waste hits your bill differently depending on mode.

Standard leaves you paying for underutilized nodes; Autopilot bills you for requested resources whether you use them or not.

### Fixing it manually doesn't scale, and specialized resources make it worse.

GPUs and high-performance disks are frequently over-provisioned for peak loads that rarely occur.

## How We Help

### Autonomous Workload Rightsizing

Sedai analyzes historical CPU and memory usage for every workload and continuously adjusts resource requests to match real demand, in both Standard and Autopilot clusters.

### Smarter Node Infrastructure (Standard)

Sedai consolidates pods onto fewer, better-utilized nodes, helps the cluster autoscaler terminate idle capacity faster, and recommends more cost-effective instance types for your node pools.

### GPU & Persistent Disk Optimization

Sedai identifies idle GPU capacity, right-sizes workloads to the most cost-effective GPU types, and flags overprovisioned or unattached Persistent Disks based on real I/O patterns.

## Stop Paying GKE Prices for Idle Capacity.

## Frequently Asked Questions

### Does Sedai work with both GKE Standard and Autopilot?

Yes. Sedai rightsizes workload requests in both modes, and additionally optimizes node infrastructure — consolidation, autoscaling, instance selection — in Standard clusters where you manage the nodes.

### How does Sedai make changes safely?

Sedai makes small, incremental adjustments and validates each one against application health metrics like latency and error rates before proceeding, rather than applying large changes all at once.

### Why does workload rightsizing matter so much in Autopilot specifically?

In Autopilot, you're billed directly for what your workloads request, not what they use, so rightsizing requests is the primary lever for reducing cost in that mode.

### Does Sedai work alongside the GKE cluster autoscaler?

Yes. By consolidating workloads and increasing node utilization, Sedai helps the cluster autoscaler identify and terminate idle nodes more effectively, rather than replacing its logic.

### Can Sedai optimize GPU workloads?

Yes. Sedai identifies idle GPU capacity, right-sizes workloads to more cost-effective GPU types, and can partition physical GPUs across multiple smaller workloads to increase density.