知识卡片

普通读书笔记卡 · 5427

长尾要看分位而非只看均值

理解等级:高级

少数请求可能远慢于平均值,扇出与多跳会增加遇到它们的机会。评估时区分均值、中位数与高分位,并为拥塞和故障变化保留观测。书中启动实例直方图是一次历史测量,不能把其95分位或倍数当当前云平台通用SLA。

这张卡片看不懂?

已有反馈 · 0 / 10 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • 云与分布式计算 · 查看原书依据

      Long tail latencies are a result of congestion or failure somewhere in the path of the service request. Many factors may contribute to congestion—server queues, hypervisor scheduling, or others—but the cause of the congestion is out of your control as a service developer. Your monitoring techniques and your strategies to achieve your required performance and availability must reflect the reality of a long tail distribution.

    参考来源

    - 位置:《Software Architecture in Practice, 4th Edition》第17章《云与分布式计算》(来源项目:book-sources;源文件:extracted/Software Architecture in Practice, 4th Edition/_epub-src/ch17.html) - 结论依据:所引连续原文给出本卡机制或取舍;条件与示例结合本节已读上下文解释。 - 原始内容:Long tail latencies are a result of congestion or failure somewhere in the path of the service request. Many factors may contribute to congestion—server queues, hypervisor scheduling, or others—but the cause of the congestion is out of your control as a service developer. Your monitoring techniques and your strategies to achieve your required performance and availability must reflect the reality of a long tail distribution.