Deploying Artificial Intelligence (AI) and Machine Learning (ML) workloads at scale has become a primary objective for modern enterprises. However, moving these data-heavy, stateful workloads into cloud native infrastructure introduces massive data bottlenecks.在大规模环境下部署人工智能（AI）和机器学习（ML）工作负载已成为现代企业的首要目标。然而，将这些数据密集型、有状态的工作负载迁移到云原生基础设施中，会带来巨大的数据瓶颈。
To help organizations navigate this fast-evolving landscape, the CNCF Technical Advisory Group for Infrastructure (TAG Infrastructure) has released its latest comprehensive white paper: Data On Kubernetes – Data Analytics and AI/ML Workloads为了帮助各组织应对这一快速发展的领域，CNCF 基础设施技术咨询小组（TAG Infrastructure）发布了最新的综合白皮书：《Kubernetes 上的数据——数据分析与 AI/ML 工作负载》。
The Challenge: Storage at the speed of AI挑战：实现 AI 速度的存储
Traditional storage architectures optimized for standard microservices fall short when tasked with feeding massive datasets into parallelized, high-performance accelerator hardware like GPUs. Infrastructure teams face unique hurdles across the data lifecycle:传统的存储架构针对标准微服务进行了优化，但在向 GPU 等并行高性能加速硬件输送海量数据集时，往往力不从心。基础设施团队在整个数据生命周期中面临着独特的挑战：
- The Small-File Trap: Datasets consisting of millions of small files put immense pressure on storage metadata servers.小文件陷阱：由数百万个小文件组成的数据集会对存储元数据服务器造成巨大压力。
- Decoupled Bottlenecks: Compute-storage disaggregation scales efficiently but can introduce heavy API call overhead and low GPU utilization rates.解耦瓶颈：计算与存储的分离虽然能实现高效扩展，但可能引入沉重的 API 调用开销，并导致 GPU 利用率低下。
- Shifting Workload Profiles: High-throughput batch training jobs require sustained data movement, whereas production inference demands low-latency, spiky request-response profiles.工作负载特征的转变：高吞吐量的批量训练任务需要持续的数据传输，而生产环境的推理任务则要求低延迟、突发性的请求-响应模式。
Key technical pillars inside the White Paper白皮书中的关键技术支柱
The white paper breaks down the cloud native AI data ecosystem into critical structural layers:该白皮书将云原生 AI 数据生态系统拆解为几个关键的结构层：
- Data Lake Houses & Vector Databases: The guide explores the merging of centralized systems into hybrid data lake houses using open formats like Apache Parquet and Iceberg. It also dives into Vector Databases (like Milvus) that handle high-dimensional embeddings for similarity searches and Retrieval-Augmented Generation (RAG).数据湖仓与向量数据库：本指南探讨了如何利用 Apache Parquet 和 Iceberg 等开放格式将集中式系统合并为混合数据湖仓。此外，它还深入介绍了处理高维嵌入以进行相似性搜索和检索增强生成（RAG）的向量数据库（如 Milvus）。
- Caching & Data Locality: To eliminate data transfer lag, the paper outlines data locality strategies, highlighting the CNCF project Fluid for orchestrating distributed caching within Kubernetes.缓存与数据局部性：为消除数据传输延迟，白皮书概述了数据局部性策略，并重点介绍了用于在 Kubernetes 内编排分布式缓存的 CNCF 项目 Fluid。
- Standardized Interfaces (CSI & COSI): Learn how the community bridges storage layers using the Container Storage Interface (CSI) for block/file storage, the Container Object Storage Interface (COSI) for object storage, and cloud-native FUSE CSI drivers.标准化接口（CSI 与 COSI）：了解社区如何利用容器存储接口（CSI）处理块/文件存储、利用容器对象存储接口（COSI）处理对象存储，以及使用云原生 FUSE CSI 驱动程序来连接存储层。
- Modern Data Pipelines: The paper maps out structural roadmaps for moving from legacy batch blocks to real-time streaming using Change Data Capture (CDC) and event streaming platforms like Apache Kafka.现代数据流水线：白皮书规划了从传统批处理模块转向利用变更数据捕获（CDC）和 Apache Kafka 等事件流平台进行实时流处理的结构化路线图。
Storage profiles across the AI lifecycleAI 生命周期中的存储配置
A major highlight of the white paper is its granular breakdown of storage footprints across three distinct phases:白皮书的一大亮点是针对以下三个不同阶段，对存储需求进行了细致的拆解：
1. Model training1. 模型训练
A long-running, throughput-oriented phase focused on maximizing GPU utilization. Storage must tolerate non-sequential access due to random data shuffling and survive massive, synchronized write bursts during checkpointing (saving model states to protect against hardware failures).这是一个侧重于吞吐量的长期运行阶段，核心在于最大化 GPU 利用率。由于存在随机数据洗牌，存储系统必须能够处理非顺序访问，并能在检查点（为防止硬件故障而保存模型状态）期间承受大规模、同步的写入突发。
2. Model inference2. 模型推理
A latency-sensitive phase characterized by spiky traffic and rapid model-loading requirements. Production systems rely heavily on advanced memory architectures like KV Caching and Prefix Caching to eliminate redundant conversational calculations.这是一个对延迟敏感的阶段，其特点是流量突发且模型加载要求极快。生产系统严重依赖 KV 缓存（KV Caching）和前缀缓存（Prefix Caching）等先进的内存架构，以消除冗余的对话计算。
3. Agentic AI (AI Agents)3. 智能体 AI（AI Agents）
Emerging AI agents introduce a complex, closed-loop iterative reasoning architecture. Their storage footprint requires short-term memory (mutable state tracking and append-only event histories), artifact repositories for intermediate code or media blobs, and long-term memory to consolidate past sessions.新兴的 AI 智能体引入了一种复杂的闭环迭代推理架构。其存储需求包括短期记忆（可变状态跟踪和仅追加事件历史）、用于中间代码或媒体块的工件存储库，以及用于整合过往会话的长期记忆。
Get involved!参与其中！
The TAG Infrastructure community welcomes your insights as we develop sustainable, practical architectural patterns for modern cloud native workloads.TAG Infrastructure 社区欢迎您贡献见解，共同为现代云原生工作负载开发可持续、实用的架构模式。
- Read the full white paper 阅读完整白皮书
- Contribute by checking out the TAG Infrastructure Charter.查看 TAG Infrastructure 章程，参与贡献。
- Join the conversation on the #tag-infrastructure channel on the CNCF Slack workspace!加入 CNCF Slack 工作区中的 #tag-infrastructure 频道，参与讨论！