Microsoft teaches NASA a lesson in timeshare optimization
Microsoft’s Azure Quantum team worked with NASA’s Jet Propulsion Laboratory (JPL) to apply quantum computing to a classical scheduling problem.
The Deep Space Network (DSN) is a collection of large antennas, based on three installations equidistant from each other – approximately 120 degrees longitude – around the world. The DSN is operated by JPL, which also operates many NASA interplanetary robotic space missions.
Space mission operations teams use the DSN command system to control the activities of their spacecraft. Commands are sent to the robotic probes as coded computer files which the machine performs in a series of actions.
The DSN also acquires, processes, decodes and distributes scientific and technical telemetry data transmitted to Earth via radio signals from spacecraft as they explore the far reaches of our solar system. It also offers capabilities to support scientific investigations that probe the nature of asteroids and the interiors of planets and moons.
However, as Nasa launches more frequent and complex missions into space, managing communications with the growing number of spacecraft is increasingly difficult.
In a blog post describing the project, Microsoft said planning demands to use space mission DSN antennas come with a lot of constraints and require intensive computing resources. “All missions require access for key communication, resulting in several hundred requests weekly when each spacecraft is visible through the antenna,” Azure Quantum senior software engineer Anita Ramanan wrote in the blog post.
According to Ramanan, the JPL scheduling task is a multivariate problem. She said the team took quantum-inspired optimization algorithms from Microsoft’s quantum computing research and ran them on classical computing architecture.
By thinking about how to solve the problem using a quantum computer, the team was able to develop a quantum-inspired algorithm that could be run on classical computer architecture. Early in the project, the Microsoft team said they logged run times of two hours or more to produce a schedule for the DSN.
Ramanan said that by applying quantum-inspired optimization algorithms with Azure Quantum, “we were able to reduce execution time to 16 minutes, and a custom solution reduced it to around two minutes.”
She added, “Schedules that are produced in minutes rather than hours allow JPL to create many candidate schedules and allow the organization to be more agile as missions and space demands increase.”