Since evaluator should make use of full system resources, we may need to: Run only one island Have a global locking mechanism, but anything running in parallel would still be consu...

DSPy Overview DSPy is a framework that lets people code their AI agent systems without any prompting (almost). It’s similar to LangChain and LangGraph, except it’s more convenient to code up and a...

MLforsystem workshop'25 The paper presents SchedCP, a control-plane framework that lets autonomous LLM agents safely optimize Linux scheduling by separating semantic reasoning (“what to optimize”)...

This paper is included in the Proceedings of the 19th USENIX Symposium on Operating Systems Design and Implementation.</h4> July 7–9, 2025 • Boston, MA, USA ### 1 Introduction Develo...

1. Introduction The paper uses Evolutionary Computation (EC) in the standard sense: keep a population of candidates → evaluate them (fitness) → generate variations → select the better ones → repea...

HotNets, November 2025 We face rapid evolution in workloads, deployment settings, and heterogeneous hardware. As such, the design space of heuristics for most problems is complex and shifting in a...

1. Introduction There lacks a study on kernel heuristics, even though heuristics is a big part of operating systems, as authors identify 68 heuristics in Linux memory management alone. The paper s...

1. Introduction The paper argues that DRAM has become too expensive as a stand-alone capacity solution because application memory demand (e.g., ML) is rising while DRAM scaling slows and costs flu...

Linux’s page cache is crucial for performance, but its default one-size-fits-all eviction policy often performs poorly, and implementing better policies in the kernel is difficult. They present ca...

Design Goals Compatibility: Solution must coexist with current kernel-based policies and swap backends; i.e., some workloads running with the default policies and others using PageFlex but shar...