# Google Persistent Disk

Sedai optimizes Google Persistent Disk autonomously, matching disk type to real I/O and clearing unattached disks — with validated, safe execution.

## Match Every Disk to Its Actual I/O

Persistent Disks are often provisioned for peak performance and left that way indefinitely — and unattached disks keep costing money long after the VM they served is gone. Sedai continuously analyzes I/O patterns and attachment state, and can recommend, assist, or autonomously apply the right fix, depending on the mode you choose.

## Disk Performance Is Provisioned Once. Usage Changes Constantly.

Two patterns quietly waste money on Persistent Disks: paying for I/O capacity you don't use, and paying for disks nothing is even using.

### High-performance disks are provisioned for peak, then forgotten.

A pd-ssd sized for worst-case I/O keeps costing the same even after usage settles into a normal, lower pattern.

### The mismatch is between disk type and actual behavior.

Provisioned performance and observed IOPS or throughput drift apart over time, and nothing flags that gap on its own.

### Unattached disks keep billing after the VM is gone.

Disks decommissioned alongside a VM are easy to forget and continue incurring cost indefinitely.

## How We Help

### Disk Performance Profiling

Sedai collects IOPS, throughput, and configuration data for every disk, using machine learning to build a performance profile based on how it's actually used.

### Recommendations or Autonomous Action, Your Choice

Over-provisioned disks and unattached disks are surfaced as recommendations in Datapilot and Copilot modes, or resolved automatically in Autopilot — the level of control is yours.

### Safe, Validated Execution in Autopilot

When running autonomously, Sedai applies changes through maintenance windows, validates system health afterward, and automatically rolls back if any issues are detected.

## Don't Pay Peak Prices for Average I/O.

## Frequently Asked Questions

### Does Sedai automatically change my disk type or delete unattached disks?

It depends on the mode you choose. In Datapilot, Sedai only recommends changes for your review. In Copilot, you get the same recommendations with more built-in support to act on them. In Autopilot, Sedai applies changes autonomously.

### What's the difference between Datapilot, Copilot, and Autopilot for Persistent Disk?

Datapilot surfaces recommendations only. Copilot builds on that with more guided support. Autopilot goes further and executes changes autonomously, validating system health afterward and rolling back automatically if issues arise.

### What data does Sedai use to evaluate a disk?

Sedai collects performance metrics like IOPS and throughput from Cloud Monitoring, along with the disk's configuration — type, size, and attachment state — to build a usage profile over time.

### What kind of disk type change might Sedai recommend or apply?

A common example is a pd-ssd disk with consistently low IOPS and throughput being changed to pd-balanced or pd-standard, reducing cost without affecting the application.

### How does Autopilot mode avoid risking application performance?

In Autopilot, Sedai executes changes through maintenance windows and validates system health immediately afterward, automatically rolling back if anything looks wrong — so the safety net doesn't depend on a human reviewing it first.