← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 7010.n · 生成模型:DeepSeek-V4.1-Flash

MaaS 的四类代价

理解等级:高级

MaaS 的代价与优势同源:管理开销——上了微服务这列车就难下,维持安全运行需要大量(好在已被充分记录与理解)开销;组织合规——标准微服务框架「免费」带来日志聚合、指标抓取与仪表盘、容器与算力元数据、把构建/发布转成真实部署的托管交付软件,同时也会带来隐私、安全标准、认证、审计、资源限额与各类迁移的变更要求;延迟预算必须留足——把调用栈外化后延迟成为不可忽略的硬约束,用户感知延迟要压在亚秒级,这会反过来约束所有协作系统的性能,还容易在筒仓式组织里形成「没有哪个团队对端到端性能负责」的盲点,使底层数据存储、语言与组织结构变得重要;分布式可用性——基于分布式微服务的架构必须稳健容忍部分失败,模型服务宕机时调用方要有合理回退。这与第 7 章「多数故障不是 ML 故障」的判断一脉相承。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • Serving · 查看原书依据

      Distributed availability Architectures built on distributed microservices must be able to tolerate partial failures robustly. The calling service must have reasonable fallbacks when the model service is down. Serving at the Edge A slightly less commonly understood serving architecture is used when a model is deployed onto edge devices (Figure 8-5).

    参考来源

    - 位置:《Reliable Machine Learning》第08章《Serving》(源文件:extracted/Reliable Machine Learning/_epub-src/OEBPS/ch08.xhtml) - 结论依据:原文明确阐述该主题,并给出定义、实践或反例。 - 原始内容:Distributed availability Architectures built on distributed microservices must be able to tolerate partial failures robustly. The calling service must have reasonable fallbacks when the model service is down. Serving at the Edge A slightly less commonly understood serving architecture is used when a model is deployed onto edge devices (Figure 8-5).