Karma - Resource Allocation for Dynamic Demands

Explore a groundbreaking approach to resource allocation for dynamic user demands in this 17-minute conference talk from OSDI '23. Delve into the innovative Karma system, which introduces memory into max-min fairness to address the limitations of classical algorithms. Learn how Karma guarantees Pareto efficiency, online strategy-proofness, and optimal fairness without future knowledge of user demands. Discover the key insights behind Karma's credit-based allocation system and its ability to reduce performance disparity across users while maintaining optimal system-wide performance. Gain valuable insights from Cornell, Columbia, and Yale University researchers as they present their findings on this cutting-edge resource allocation technique, backed by both theoretical proofs and empirical evaluations using production workloads.

OSDI '23 - Karma: Resource Allocation for Dynamic Demands