Noise of Ethernet technology development

Ethernet technology has been passive learning for a long time. Nowadays, it is necessary to take the initiative. Passive depends on a series of experience values. For example, the STPS duration is 15 seconds, the information obtained by all vswitches within 30 seconds will be in a consistent state. If you do not have to wait for so long according to the active principle, can you just ask?
The Development of Traditional Ethernet has gone through several stages:
1. CSMA/CD bus half duplex phase:Bus topology.
2. Hub half duplex phase:The topology is changed to a star topology.
3. Learning switch stage:It is still a star topology. Due to the increasing importance of the bearer service, in order to provide redundancy, the topology should be a hybrid topology with a ring mesh.
4. logic tree stage:Since Ethernet is actually a broadcast network, in a redundant Ethernet environment, in order to avoid broadcast storms, what we need is a logically tree-like topology, STP/RSTP /...
However, these are no longer suitable for today's environments. stps slow convergence and rstps partially solves this problem. However, in many cases, STPS algorithms are rolled back to the underlying causes, that is, Ethernet has never had a good "routing" function, and has always been passively converging, with few active behaviors, although RSTP adds the feature of active negotiation to shorten the convergence time to less than 1 second in some cases, this is not thorough. The switch learning mechanism is also the same. Sometimes it will be rolled back to broadcast, so the Ethernet switch learning mechanism or RSTP is only a Traditional Ethernet optimization at best, there is no qualitative leap.

Ethernet has always been a winner, replacing many LAN technologies and even killing the technology in its infancy. However, it is also true that sometimes it is fixed. Today's WAN technology basically solves the link layer routing problem. If the link layer technology itself does not have a routing mechanism, it encapsulates an X.5 layer, and the ATM switch implements the link layer routing well, by establishing a virtual channel to build a forwarding link for the upper layer of data, or even layer-3 technology, we all know that IP Forwarding is very slow (Cisco CEF technology is not considered for the time being ), as a result, the MPLS layer 2.5 is implemented. A label ing table is constructed in the MPLS switch through the Label Distribution Protocol in advance, then, the IP datagram is encapsulated into the MPLS header to implement Fast Forwarding Based on tags rather than slow route lookup.
Ethernet is not visible, maybe it is not worth implementing. After all, Ethernet is already working well enough. I think it was a cost-effective solution, otherwise, the twisted pair cable will not replace many cables with far better features. It is time to change this because Ethernet is not only serving an office area, but also carrying a large amount of information and applications out of cloud endpoints, it may be part of a cluster, or it may be the core of Iot control... the disadvantages of STP and other technologies are apparent. An active link layer routing protocol must be designed to make the optimal path selection and physical redundancy compatible. According to the design principles of layered protocols, this routing technology should not rely on any upper layer technology. However, to be compatible with standard 802.3 Ethernet, a new layer 1.5 encapsulation is required, we do not want to design a new link layer, but to design a technology that enables Ethernet to achieve fast convergence, proactive route computing, and optimal path forwarding over Ethernet frames. Based on the above, mature technologies such as VPN, GRE, MPLS cannot become candidates. In fact, Cisco's fabricpath is developed in this context. In the end, Ethernet completed its final blow, making it perfect for the cloud environment, so it entered a new stage:
5. Fast forwarding by optimal routing:In terms of technical implementation, fabricpath is just an example. There are many similar technologies in advance. fabricpath uses the is-is routing protocol to calculate the optimal route. With the best route, no block port is required, and all the ports are forwarded. It is determined by the route computing result that the port cannot be taken away!
Now that we know the imperfect nature of STP technology, it must block some ports to physically cut off some paths. It is precisely because of Ethernet broadcasting that Ethernet itself is fully broadcast, the emergence of a Learning switch limits most broadcasts. However, when this learning mechanism reaches its limit (for example, no ing is found in the Mac table), frames are still broadcast, in addition, Ethernet has a broadcast MAC address with a full value of 1. Such frames must be broadcast in any way and are not involved in the learning process. This silly and tough broadcast behavior makes STP have to block the Port hard. After the link layer routing problem is solved, the next step is to solve the broadcast problem. The TTL mechanism is a good choice and also the fabricpath choice, we can also use the rip anti-ring method as an optional solution. Considering that the Ethernet scale has an predictable maximum value, we can adopt the toxicity reversal principle, or horizontal segmentation with time-out latency is acceptable, and the trigger update mechanism is also an option, which is somewhat similar to the active report/response mechanism of RSTP. In short, technical means can be integrated, but the principle of solving the problem is unique.