惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

罗磊的独立博客
爱范儿
爱范儿
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
博客园_首页
博客园 - 叶小钗
酷 壳 – CoolShell
酷 壳 – CoolShell
Apple Machine Learning Research
Apple Machine Learning Research
云风的 BLOG
云风的 BLOG
量子位
博客园 - 三生石上(FineUI控件)
Stack Overflow Blog
Stack Overflow Blog
小众软件
小众软件
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
V
V2EX
人人都是产品经理
人人都是产品经理
V
Visual Studio Blog
Jina AI
Jina AI
L
LangChain Blog
M
MIT News - Artificial intelligence
MongoDB | Blog
MongoDB | Blog
Last Week in AI
Last Week in AI
Martin Fowler
Martin Fowler
WordPress大学
WordPress大学

cs.DC updates on arXiv.org

DUAL-BLADE: Dual-Path NVMe-Direct KV-Cache Offloading for Edge LLM Inference Progressive Semantic Communication for Efficient Edge-Cloud Vision-Language Models Efficient, VRAM-Constrained xLM Inference on Clients Folding Tensor and Sequence Parallelism for Memory-Efficient Transformer Training & Inference DORA: A Scalable Asynchronous Reinforcement Learning System for Language Model Training AMMA: A Multi-Chiplet Memory-Centric Architecture for Low-Latency 1M Context Attention Serving RaMP: Runtime-Aware Megakernel Polymorphism for Mixture-of-Experts Spark Policy Toolkit: Semantic Contracts and Scalable Execution for Policy Learning in Spark Internet of Everything in the 6G Era: Paradigms, Enablers, Potentials and Future Directions PolyKV: A Shared Asymmetrically-Compressed KV Cache Pool for Multi-Agent LLM Inference A Survey on Split Learning for LLM Fine-Tuning: Models, Systems, and Privacy Optimizations ITAS: A Multi-Agent Architecture for LLM-Based Intelligent Tutoring Latency and Cost of Multi-Agent Intelligent Tutoring at Scale TACO: Efficient Communication Compression of Intermediate Tensors for Scalable Tensor-Parallel LLM Training FreeScale: Distributed Training for Sequence Recommendation Models with Minimal Scaling Cost CommFuse: Hiding Tail Latency via Communication Decomposition and Fusion for Distributed LLM Training A Taxonomy and Resolution Strategy for Client-Level Disagreements in Federated Learning Usable Agent Discovery for Decentralized AI Systems Cloud to Edge: Benchmarking LLM Inference On Hardware-Accelerated Single-Board Computers Data-Free Contribution Estimation in Federated Learning using Gradient von Neumann Entropy Shard the Gradient, Scale the Model: Serverless Federated Aggregation via Gradient Partitioning Promoting Simple Agents: Ensemble Methods for Event-Log Prediction GraphLeap: Decoupling Graph Construction and Convolution for Vision GNN Acceleration on FPGA AGNT2: Autonomous Agent Economies on Interaction-Optimized Layer 2 Infrastructure FedSIR: Spectral Client Identification and Relabeling for Federated Learning with Noisy Labels Stream-CQSA: Avoiding Out-of-Memory in Attention Computation via Flexible Workload Scheduling A Delta-Aware Orchestration Framework for Scalable Multi-Agent Edge Computing Federated Learning over Blockchain-Enabled Cloud Infrastructure Optimal Routing for Federated Learning over Dynamic Satellite Networks: Tractable or Not? Sherpa.ai Privacy-Preserving Multi-Party Entity Alignment without Intersection Disclosure for Noisy Identifiers
The Ghost in the Datacenter: Link Flapping, Topology Know...
[Submitted on 4 Mar 2026 (v1), last revised 9 Sep 2026 (this ver · 2026-03-04 · via cs.DC updates on arXiv.org

View PDF HTML (experimental)

Abstract:A link changes state faster than knowledge of that change can reach every component that acts on it. During the interval in between, some observer operates on a topology that no longer exists. We call that mismatch a ghost: for observer i, a ghost exists whenever the connectivity state on which i acts differs from the connectivity actually available to it. Ghosts appear wherever this delay does -- PCIe and UCIe, NVLink and NVSwitch, Ethernet and Thunderbolt, IP and BGP. What the mechanisms we survey have in common is narrower than a shared ancestry: in each, failure state is at some point inferred from absence, elapsed time, probing, or eventual convergence, and absence alone does not distinguish delay from loss, partition, or endpoint failure.
We survey the scale of the problem with production data from Meta, ByteDance, Google and Alibaba, and reproduce an industry scaling projection -- attributed, and not our own measurement -- under which a fabric of more than ten million optical links sees an aggregate flap roughly every 48 seconds. The mitigations surveyed reduce the interval, often substantially, but each retains a period in which failure state is inferred rather than established.
We then examine Open Aethernet (OAE), whose distinctive move is asymmetric: a completed transaction yields positive, mutually held evidence, while absence of completion is not treated as identification of a failure but recorded as an explicit IN-DOUBT state, resolved by corroboration where the topology permits. We are explicit about what this does not claim. OAE does not eliminate uncertainty; it makes it first-class. It does not repair a hard partition; detection is not repair; its detector latency is uncharacterised; triangle coverage is assumed rather than proved; we report no OAE measurement; and we do not claim common knowledge in the epistemic-logic sense.

Submission history

From: Paul Borrill [view email]
[v1] Wed, 4 Mar 2026 05:12:40 UTC (21 KB)
[v2] Wed, 9 Sep 2026 03:57:32 UTC (28 KB)