Skip to main content
Suspend and resume is coming soon to boxd. This page describes how it will work. The underlying technology is already built — boxd is based on Ignition, an open-source microVM orchestrator by Azin that powers sub-10ms cold starts.

​
The idea

Today, boxd machines run continuously. With suspend and resume, machines can be frozen in place — all memory, processes, and network state preserved — and woken up in under 10 milliseconds when traffic arrives. This means you can have dozens of machines that cost nothing when idle and wake instantly when needed.

​
How it works

Ignition uses a snapshot-based approach to eliminate initialization overhead:
  1. Run and snapshot. Your machine boots, your app initializes, and Ignition captures the full VM state — memory, CPU registers, device state — into a snapshot.
  2. Suspend when idle. When no traffic is flowing and no sessions are active, the machine suspends. The snapshot is retained in memory or on disk.
  3. Resume on demand. When a request arrives at name.boxd.sh, the proxy detects the machine is suspended and triggers a restore. The VM resumes from the snapshot — no boot, no init, no application startup. It picks up exactly where it left off.
The restore happens in under 10 milliseconds. From the client’s perspective, the machine was always running.

​
Snapshot strategies

Ignition can capture the snapshot at different points:
  • After first listen — snapshot once the app opens its first network socket (default)
  • After Nth listen — wait for multiple sockets (e.g., app + database)
  • After specific port — snapshot when a particular port is ready
  • After userspace ready — snapshot once the kernel hands off to userspace
  • Manual — your app signals when it’s ready
The modified Linux kernel detects these triggers automatically. No code changes needed in your application.

​
State retention

Snapshots can be retained in two ways:
  • In-memory — fastest resume (~1ms), uses RAM while suspended
  • On-disk — slightly slower resume (~5-10ms), no RAM cost while suspended

​
What this enables

  • Pay for what you use. Machines that serve occasional traffic suspend between requests.
  • Massive parallelism. Spin up hundreds of machines, let them suspend when idle, wake on demand.
  • Instant APIs. Deploy a service that cold-starts faster than a DNS lookup.
  • Agent workspaces. An agent’s VM suspends when the conversation ends and resumes instantly when the next one starts.

​
Learn more

Ignition is open source. Read the code and architecture:

​
Reference

CLI reference

Every command and its flags.

Resources and limits

Quotas, CPU, memory, and disk.