Spirit: Fair Allocation of Interdependent Resources in Remote Memory Systems

Fair allocation when resources are interdependent — remote memory bandwidth, capacity, and compute interact

Featured image

Venue: SOSP 2025

Topic: Remote memory systems disaggregate memory from compute, but fairly allocating interdependent resources (bandwidth, capacity, compute) across tenants is non-trivial. Spirit proposes a fair allocation mechanism for such systems.


Summary

Remote memory systems (e.g., CXL-based disaggregated memory) expose multiple interdependent resources: memory capacity, memory bandwidth, and local compute. Standard fair allocation (e.g., max-min fairness) fails when resources are interdependent — allocating one resource constrains what another can do for a tenant. Spirit addresses fair allocation of interdependent resources in remote memory systems, ensuring no tenant is unfairly starved.


Background

Remote memory systems

Interdependence


Key Idea


Design

(Based on meeting discussion)


Meeting Notes