Introduction

Architect enables any application to run continuously on spot instances with zero downtime and zero data loss during preemption events - all without requiring code modifications. By making spot instance preemption irrelevant, Architect unlocks immediate 75%+ infrastructure cost savings while maintaining production-grade reliability.

Key Features

• Zero Downtime Migration: Continuously capture and transfer application state between compute resources within the preemption window, maintaining active network connections without interruption
• No Code Changes Required: Deploy existing applications on spot instances without implementing checkpoint/restore logic or other modifications
• Complete State Preservation: Retain all application state during migrations, including CPU, memory, and GPU state
• Cross-Environment Compatibility: Seamlessly move workloads between different CPU architectures, GPU types, cloud providers, and regions
• Universal Workload Support: Run any application on spot instances - from ML training and inference to databases, game servers, and data pipelines

How Architect Works

Unlike traditional checkpoint-based approaches, Architect takes a fundamentally different approach to handling spot instance preemption:

1. Continuous State Capture: Efficiently serialize and store application state to passive storage (EBS, S3, etc.)
2. Preemption Detection: Receive and respond to cloud provider preemption notices automatically
3. Resource Acquisition: Identify and secure target compute resources across available providers
4. Live Migration: Stream application state to the new instance, combining previously captured state with live changes
5. Traffic Management: Automatically reroute network connections without dropping active sessions

The entire process completes within the preemption window, making instance termination irrelevant to your running applications.

Going Beyond Spot Instances

While immediate cost savings through spot instance usage is a primary benefit, Architect's capabilities extend far beyond:

• Deploy truly portable multi-cloud infrastructure
• Perform zero-downtime hardware upgrades
• Optimize geographic placement for latency
• Eliminate downtime from resource failures
• Move compute closer to users on demand