Searcharxiv⌕ Search

arXiv subjects

Yanpeng Hu

Publications and source records attributed to Yanpeng Hu.

10 recordsLinked to original sources
arXiv2026

HBFSim: Fast and Faithful Simulation of High-Bandwidth Flash Under Real GPU Execution

Serving a large language model (LLM) is limited by memory capacity. High-Bandwidth Flash (HBF) stacks NAND flash inside the accelerator package, one tier below high-bandwidth memory (HBM); the specification was published on August 3, 2026, and the first inference devices are expected to sample in early 2027. Decisions about capacity and data placement cannot wait for silicon. No existing method settles those decisions: a storage simulator replaying a recorded access sequence never executes the workload, a GPU simulator does not run the real compute kernels, and a cycle-accurate simulator cannot finish one LLM inference run. We present HBFSim, the first evaluation platform that applies HBF timing, capacity, and thermal effects to a real inference workload while that workload executes on a real GPU. HBFSim rewrites PTX, the intermediate code NVIDIA's compiler emits, and gates kernel launches; issue is separated from consumption, so real hardware supplies the computation that hides an access. Timing comes from measurements of a real device rather than a parameter sheet, and junction temperature sets both the rate HBF sustains and the retention deadline that forces refresh writes. HBFSim matches the measured device exactly at all six calibration breakpoints, with zero unsafe launches, and an unmodified vLLM 0.15.1 serving Qwen3-30B-A3B returns the token identifiers of the uninstrumented baseline. The device fast path serves the same Qwen3-30B-A3B case in 2s against 44s on the detailed reference path, 20.8x faster. Before HBF parts sample, HBFSim lets a designer measure a capacity or placement decision under a real workload instead of assuming one.

cs.AR↗
arXiv2026

When Can Agents Safely Checkpoint, Fork, Restore, and Merge? Exact Checking for Execution Edits

Agent runtimes can Checkpoint an execution, Fork it, Restore a checkpoint, or Merge branches without restarting a task. We call these operations execution edits, with Checkpoint recording the current execution for later use and Fork, Restore, and Merge changing what the Agent will do next. An execution edit cannot undo an earlier authorization or a tool request already sent. An unsafe edit can therefore authorize the same tool action twice, discard a result the task still requires, or conflict with a call that began before the edit. The Agent is untrusted, so the runtime uses its execution record to determine which past actions an edit must account for and which required results it must preserve to keep the subsequent execution safe. Yet existing Agent systems support such operations without deriving what each edit must preserve from the running execution, whereas prior methods for computing safe behavior take that requirement as input. We give an algorithm that decides exactly whether an edit is safe. It returns all safe ways to continue, or proves that none exists. To make this decision, the algorithm lists every way the task can finish without violating policy. It removes any way that could make a still-required result impossible to finish later. If none remain, it returns a checkable proof that no safe implementation exists. Otherwise, the remaining ways describe exactly what the runtime may allow. Our formal results cover Checkpoint and the six forms of Fork, Restore, and Merge, together with extensions, atomic enforcement, and the information every exact checker needs. Lean mechanizes the finite checker and runtime invariant, and tests validate all six edit forms. The source code, Lean proofs, and executable tests are available in the public GitHub repository at https://github.com/eunomia-bpf/agent-check-restore-safety.

cs.PL↗
arXiv2026

ReFlux: Reversible Compute Placement for CXL-Enabled Storage

Static offload to computational storage devices proves brittle because device-side processors throttle under sustained thermal load, while opaque, vendor-specific interfaces inflate adoption costs so severely that no computational storage platform has achieved broad deployment; to address this, we argue that storage-side compute should be reversible, allowing individual pipeline stages to migrate between host and device at runtime beneath standard interfaces that require zero application modification. We present ReFlux, which realizes this principle on CXL SSDs by decomposing I/O-path logic into migratable storage actors compiled to WebAssembly, with actors sharing state through coherent CXL.mem regions so that only lightweight control state, roughly 8 KB, moves during migration, while a thermal-aware scheduler triggers per-stage drain-and-switch when device temperature or queue pressure rises, offloading compute-intensive actors to the host while leaving I/O-bound stages on the device. In our evaluation on an FPGA-based CXL SSD prototype and two production CSDs, ReFlux sustains over twice the throughput of thermally throttled CSDs under 30-minute sustained writes, delivers three to four times the inference throughput under KV-cache pressure, and reduces host CPU utilization by 65% through MWAIT-based notification.

cs.OS↗
arXiv2025

gpu_ext: Extensible OS Policies for GPUs via eBPF

Performance in modern GPU-centric systems increasingly depends on resource management policies, including memory placement, scheduling, and observability. However, uniform policies typically yield suboptimal performance across diverse workloads. Existing approaches present a tradeoff: user-space runtimes provide programmability and flexibility but lack cross-tenant visibility and fine-grained control of hardware resources; meanwhile, modifications to the OS kernel introduce significant complexity and safety risks. To address this, we argue that the GPU driver and device layer should provide an extensible OS interface for policy enforcement. While the emerging eBPF technology shows potential, directly applying existing host-side eBPF is insufficient because they lack visibility and control into critical device-side events, and directly embedding policy code into GPU kernels could compromise safety and efficiency. We propose gpu_ext, an eBPF-based runtime that treats the GPU driver and device as a programmable OS subsystem. gpu_ext extends GPU drivers by exposing safe programmable hooks and introduces a device-side eBPF runtime capable of executing verified policy logic within GPU kernels, enabling coherent and transparent policies. Evaluation across realistic workloads including inference, training, and vector search demonstrates that gpu_ext improves throughput by up to 4.8x and reduces tail latency by up to 2x, incurring low overhead, without modifying or restarting applications

cs.OS↗
arXiv2025

Towards Agentic OS: An LLM Agent Framework for Linux Schedulers

Operating system schedulers suffer from a fundamental semantic gap, where kernel policies fail to understand application-specific needs, leading to suboptimal performance. We introduce SchedCP, the first framework that enables fully autonomous Large Language Model (LLM) agents to safely and efficiently optimize Linux schedulers without human involvement. Our core insight is that the challenge is not merely to apply a better LLM, but to architect a decoupled control plane that separates the AI's role of semantic reasoning ("what to optimize") from the system's role of execution ("how to observe and act"), thereby separating the optimization problem into two stages: goal-inference and policy-synthesis. Implemented as Model Context Protocol(MCP) server, SchedCP provides a stable interface with three key services: a Workload Analysis Engine, an evolving Scheduler Policy Repository, and an Execution Verifier that validates all AI-generated code and configure before deployment with static and dynamic analysis. We demonstrate this architecture's power with sched-agent, a multi-agent system that autonomously analyzes workloads, synthesizes custom eBPF scheduling policies, and deploys them via the sched\_ext infrastructure. Our evaluation shows that SchedCP achieves up to an 1.79x performance improvement, and a 13x cost reduction compared to naive agentic approaches, all while maintaining high success rate. By bridging the semantic gap, SchedCP democratizes expert-level system optimization and represents a step towards creating truly self-optimizing, application-aware operating systems. The code is open-sourced in https://github.com/eunomia-bpf/schedcp

cs.AI↗
arXiv2025

AgentSight: System-Level Observability for AI Agents Using eBPF

Modern software infrastructure increasingly relies on LLM agents for development and maintenance, such as Claude Code and Gemini-cli. However, these AI agents differ fundamentally from traditional deterministic software, posing a significant challenge to conventional monitoring and debugging. This creates a critical semantic gap: existing tools observe either an agent's high-level intent (via LLM prompts) or its low-level actions (e.g., system calls), but cannot correlate these two views. This blindness makes it difficult to distinguish between benign operations, malicious attacks, and costly failures. We introduce AgentSight, an AgentOps observability framework that bridges this semantic gap using a hybrid approach. Our approach, boundary tracing, monitors agents from outside their application code at stable system interfaces using eBPF. AgentSight intercepts TLS-encrypted LLM traffic to extract semantic intent, monitors kernel events to observe system-wide effects, and causally correlates these two streams across process boundaries using a real-time engine and secondary LLM analysis. This instrumentation-free technique is framework-agnostic, resilient to rapid API changes, and incurs less than 3% performance overhead. Our evaluation shows AgentSight detects prompt injection attacks, identifies resource-wasting reasoning loops, and reveals hidden coordination bottlenecks in multi-agent systems. AgentSight is released as an open-source project at https://github.com/agent-sight/agentsight.

cs.OS↗
arXiv2023

bpftime: userspace eBPF Runtime for Uprobe, Syscall and Kernel-User Interactions

In kernel-centric operations, the uprobe component of eBPF frequently encounters performance bottlenecks, largely attributed to the overheads borne by context switches. Transitioning eBPF operations to user space bypasses these hindrances, thereby optimizing performance. This also enhances configurability and obviates the necessity for root access or privileges for kernel eBPF, subsequently minimizing the kernel attack surface. This paper introduces bpftime, a novel user-space eBPF runtime, which leverages binary rewriting to implement uprobe and syscall hook capabilities. Through bpftime, userspace uprobes achieve a 10x speed enhancement compared to their kernel counterparts without requiring dual context switches. Additionally, this runtime facilitates the programmatic hooking of syscalls within a process, both safely and efficiently. Bpftime can be seamlessly attached to any running process, limiting the need for either a restart or manual recompilation. Our implementation also extends to interprocess eBPF Maps within shared memory, catering to summary aggregation or control plane communication requirements. Compatibility with existing eBPF toolchains such as clang and libbpf is maintained, not only simplifying the development of user-space eBPF without necessitating any modifications but also supporting CO-RE through BTF. Through bpftime, we not only enhance uprobe performance but also extend the versatility and user-friendliness of eBPF runtime in user space, paving the way for more efficient and secure kernel operations.

cs.OS↗
arXiv2023

Pome: Parallelizing I/Os and Computations for Efficient LSM-tree-based Data Storage

CPU computations and I/O operations are fundamental to data storage systems. Storage systems conduct computations with their user threads, such as sorting data for orderliness. They handle I/Os mainly through system calls (syscalls) including file write, read, and fsync, which the OS's kernel threads perform with storage devices. Today, LSM-tree-based storage systems are widely deployed in production environments. Compaction is an essential operation that LSM-tree employs to maintain its tiered tree-like structure by re-sorting and re-storing data through computations and I/Os,respectively. In this paper, we first overhaul the procedure of a compaction. We find that computations and I/Os execute in sequential order. After re-sorting data, the user thread waits for a kernel thread to complete file write and fsync I/Os. These costly synchronous I/Os create a severely long critical path that affects the performance of LSM-tree. To address this issue, we propose parallelizing I/Os and computations for efficient LSM-tree-based data storage (Pome). Pome decouples computations from I/Os within each compaction by referring to its new protocol that moves I/O operations out of the critical path. To this end, it leverages the io_uring to perform asynchronous I/Os. Furthermore, regarding the potential I/O congestion caused by accelerated compactions, Pome incorporates an adaptive I/O rate limiter to achieve smooth execution. We prototype Pome on top of RocksDB. Experimental results demonstrate that Pome significantly improves the performance of RocksDB and outperforms several state-of-the-art LSM-tree variants.

cs.DB↗
arXiv2021

Accelerating LSM-Tree with the Dentry Management of File System

The log-structured merge tree (LSM-tree) gains wide popularity in building key-value (KV) stores. It employs logs to back up arriving KV pairs and maintains a few on-disk levels with exponentially increasing capacity limits, resembling a tiered tree-like structure. A level comprises SST files, each of which holds a sequence of sorted KV pairs. From time to time, LSM-tree redeploys KV pairs from a full level to the lower level by compaction, which merge-sorts and moves KV pairs among SST files, thereby incurring substantial disk I/Os. In this paper, we revisit the design of LSM-tree and find that organizing multiple KV pairs in an SST file entails the heavyweight redeployment of actual KV pairs in a compaction. Accordingly we revolutionize the organization of KV pairs by transforming an SST file of KV pairs to an SST directory, in which each KV pair makes into an independent KV file with the key and value as filename and main file contents, respectively. Moving KV pairs in a compaction converts to transferring directory entries (dentrys), which causes concretely fewer disk I/Os. This is the essence of our design named DeLSM. We build a prototype of DeLSM on LevelDB and evaluation results show that it significantly outperforms the state-of-the-art LSM-tree variants in different dimensions.

cs.DS↗