Policy comes primarily out of business decisions, and business decisions should be close to the business, not the topology. Hence, policy, or least some element of policy, is often best done when centralized.
Topology and reachability,
however, are grounded in what should be the only source of truth about
the state of the network, the network itself. Therefore, it makes sense that decisions related to the topology and reachability, from detection to reaction, should be kept close to the network itself; hence, topology and reachability decisions should trend toward being decentralized
Label则可以很好的用作TraceID，这样全网的监控不就变得容易了么，某家在前年就实现了一个在路由器上扩展容器做全网监控的软件，具体不多说，说一些公开的，阿里云也有类似的东西会在今年的SIGCOMM中呈现，那就是<VTrace: Automatic Diagnostic System for Persistent Packet Loss in Cloud-Scale Overlay Network>：