知识卡片
USL解释节点越多越慢
通用可扩展性定律用C(p)=p/[1+σ(p−1)+κp(p−1)]估算相对单节点吞吐:p是节点数,σ表示争用开销,κ表示节点间保持状态一致的开销。两项都为零才接近线性;同步代价随节点数增长可吞掉收益。按公式代入p=10、σ=κ=0.1,C约0.917,即十节点反而略慢于一个节点。这里是模型演算,不是压测结果。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
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
- 高可用与扩展 · 查看原书依据
Coordination between nodes and any data that is kept synced brings down the performance quickly. For example, using 0.1 for both contention and coherence, scalability with ten nodes is 0.9, which means a ten-node system runs slower than the single-node system.
参考来源
- 位置:《Software Architecture and Decision-Making》第09章《高可用与扩展》(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/ch09.xhtml)
- 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:Coordination between nodes and any data that is kept synced brings down the performance quickly. For example, using 0.1 for both contention and coherence, scalability with ten nodes is 0.9, which means a ten-node system runs slower than the single-node system.