知识卡片

普通读书笔记卡 · 2142.v.2

尾延迟要逐因消除

理解等级:高级

远高于常态的延迟通常出现在P99或P99.9以上,可能来自GC、突发请求和其他进程资源争用。降低平均到达率不能保证消除这些特殊事件,应逐个定位来源并处理。强尾延迟保证往往消耗额外资源、增加复杂度,只在业务真正需要时投入相应成本。

相关:[[延迟预算约束接纳速率]]。

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

    • 理解性能的心智模型 · 查看原书依据

      We can’t use the approaches we discussed to improve tail latencies, which are caused by events that are different from the norm, such as GC, a burst of requests, and resource sharing with other processes. To handle tail latencies, we need to identify each and then eliminate them. Although effective, the techniques to handle tail latencies are expensive, waste resources, and add more complexity. We should therefore design systems for strong latency guarantees only when necessary.

    参考来源

    - 位置:《Software Architecture and Decision-Making》第3章《理解性能的心智模型》(来源项目:book-sources;源文件:extracted/Software Architecture and Decision-Making Leveraging Leadership, Technology, and Product Management to Build Great Products… (Srinath Perera)/_epub-src/ch03.xhtml) - 结论依据:原文的定义、比较或例子支持“远高于常态的延迟通常出现在P99或P99.9以上,可能来自GC、突发请求和其他进程资源争用”;正文据此说明用途和边界。 - 原始内容:We can’t use the approaches we discussed to improve tail latencies, which are caused by events that are different from the norm, such as GC, a burst of requests, and resource sharing with other processes. To handle tail latencies, we need to identify each and then eliminate them. Although effective, the techniques to handle tail latencies are expensive, waste resources, and add more complexity. We should therefore design systems for strong latency guarantees only when necessary.