Performing Live Migrations of VMs

5 hours ago

Sailboxes offer a cost-effective pricing model, charging only for active resource usage per second, making long-horizon agent workflows scalable.
Live migration technology enables Sailboxes to move VMs between hosts seamlessly, with minimal downtime, supporting ephemeral compute and load balancing.
A network proxy handles open connections during migrations, spooling packets to maintain connectivity and providing additional features like whitelists and credential injection.
P2P migrations use userfaultfd for efficient memory transfer, paging in data on demand and using larger memory chunks to reduce latency.
Elastic memory consumption via virtio-mem allows Sailboxes to dynamically adjust memory usage, optimizing resource allocation and speeding up migrations.
Sailboxes charge $0.015/vCPU-hour, significantly cheaper than competitors, with billing based solely on active usage for cost-effectiveness.
Trade-offs include occasional latency spikes during migrations, but these are infrequent and manageable for long-horizon agents.
Sailboxes are now generally available, with ongoing infrastructure improvements planned for fleet utilization and caching systems.

Hasty Briefsbeta