LEO satellites are becoming a new cloud frontier — rich compute riding on the same fleets that deliver global connectivity. But the data center moves at 27,000 km/h, and orchestrators built for terrestrial clouds quietly assume the ground doesn't. Krios is the orchestration system that closes that gap.
Terrestrial orchestrators assume the servers stay put.
Deploy an app to a satellite and the satellite leaves. Within minutes, the node serving your region is over a different continent. State-of-the-art orchestrators force a brutal workaround: pin a copy of the application to every satellite that will ever pass overhead — hundreds of copies, paid for in latency, bandwidth, and money.
LEO zones: say where, not which.
Krios is designed around a novel abstraction: application providers declare the zones where their application should be available — a region on Earth, not a serial number in orbit. Krios maps each zone to whichever satellites are overhead, now and next.
Handoffs the application never feels.
Under the abstraction, Krios provides the system support that makes it real: proactive scheduling against known orbital mechanics, migration before a satellite sets, and continuity mechanisms that keep the application available in its zone without interruption. Mobility becomes the platform's problem — where it belongs — instead of every developer's.
The scheduler under the ToDare cloud.
A compute cloud you can rent by the minute needs orchestration that makes orbit look boring. Krios is that layer in the ToDare platform: customers buy availability over a footprint, and the mesh handles the rest.
@inproceedings{krios2024,
author = {Bhosale, Vaibhav and Gavrilovska, Ada and Bhardwaj, Ketan},
title = {Krios: Scheduling Abstractions and Mechanisms for Enabling a LEO Compute Cloud},
booktitle = {ACM Symposium on Cloud Computing (SoCC '24)},
year = {2024},
note = {Best Paper Award}
}