That's a crossover we never saw coming - Facebook borrows the Linux scheduler from the Steam Deck to help run its data centers
Meta is testing a Steam Deck Linux scheduler on servers, betting that latency-driven design scales beyond gaming into production infrastructure.
- Meta is deploying a Steam Deck Linux scheduler on parts of its production servers
- SCX-LAVD was originally designed to reduce latency in handheld gaming systems
- Large server machines exposed weaknesses in traditional Linux scheduling behavior
Meta has revealed that it is deploying a Linux CPU scheduler originally designed for Valve’s Steam Deck across parts of its production server fleet.
The scheduler, known as SCX-LAVD, was created to reduce latency in handheld gaming systems, yet Meta engineers now say it can address scheduling inefficiencies on large server machines.
This announcement is interesting because it links consumer gaming hardware directly to hyperscale infrastructure decisions.
According to Meta’s engineers, the motivation was not novelty but persistent scheduling limitations on modern servers.
Large machines with dozens or hundreds of CPU cores exposed weaknesses in traditional Linux scheduling behavior.
Shared scheduling queues became congested, pinned threads interfered with unrelated workloads, and network-heavy services distorted fairness calculations.
These problems appeared regardless of whether workloads were running on SSD-backed systems or interacting with cloud storage layers.
Sign up to the TechRadar Pro newsletter to get all the top news, opinion, features and guidance your business needs to succeed!
SCX-LAVD operates using the sched_ext framework, which allows alternative schedulers to plug into the Linux kernel without permanent modification.
Instead of relying on fixed priorities, the scheduler observes task behavior and dynamically estimates which tasks are latency sensitive.
Meta engineers explained that this approach required adjustments when scaled to server-class hardware, particularly to handle cache locality and cores saturated by network interrupts.
In some cases, the system treated certain cores as effectively slower to preserve overall balance.
A key point emphasized by Meta is that these changes did not require per-service tuning or manual priority assignment.
The scheduler adapts based on observed behavior rather than predefined rules.
This characteristic matters in a data center environment where workloads change frequently and manual tuning becomes expensive to maintain.
Meta suggests this reduces complexity across fleets running messaging systems, caching layers, and backend services.