paper005：Measurements As First-class Artifacts

INFOCOM`2019

5. 关键部分（用来记录本篇文章你认为关键的地方，标注好内容和页码）
6. 研究方法（可以看出此类问题，大家都是采用一个什么样的方法去解决的）
7. 存在问题（如前所述）
8. 产生原因（有些文献可能没有，没有即不写）
9. 解决方法（如前所述）
10. 创新点
11. 下一步工作/不足之处（别人论文里的不足可能是你idea的来源）

设计合适的意图原语应该符合以下原则：
（1）这些原语不仅可以表示常见测量指标，而且可以为以后提出的新的测量指标提供复用的可能性，进而提升原语的可拓展性；
（2）这些原语应该处于一个合适的抽象级别，既可以处于较低抽象级别从而方便实现多种测量任务，体现意图原语的灵活性，又可以处于较高的抽象级别从而简化用户操作，体现基于意图的方法可以降低工作量的特点；
（3）这些原语应该考虑可编程交换机的特性，充分利用可编程交换机的优势。

A core tenet of our work is that many of the common network measurement tasks can be expressed by composing primitives that can be supported by current and future programmable forwarding elements. But what is a good primitive?

1. Functions that are routinely applicable for a range of measurement needs is a candidate given its potential for reuse.
2. Functions also need to be composable to express more complex tasks. They should be sufficiently low-level to be broadly applicable but sufficiently high-level to reduce effort.
3. However, modern data rates of high-speed networks impose stringent per-packet processing requirements. Thus, primitives should have low processing and state complexity so they can be implemented in programmable forwarding elements.