Bounded Resource Reclamation

Resource allocation is well-studied in operating systems, but resource reclamation remains largely underexplored. This paper investigates the impact of unpredictable resource reclamation latency on system behavior, particularly in resource- and time-constrained environments like Open-RAN and serverless functions. We study scenarios of high reclamation times across various systems. Under adverse conditions, reclaiming resources can delay process termination by multiple seconds on both Linux and the L4Re microkernel. We propose a design for accounting and bounding resource reclamation latency to enable predictable system operation and mitigate potential denial-of-service scenarios. We also advocate for optimizing for the case of bulk reclamation — reducing worst-case reclamation latency by multiple orders of magnitude.