Under the SDN wave, where will traditional routing technology go?

introduction

"All martial arts come from Shaolin". In various martial arts novels, Shaolin Temple and "Shaolin School" are the most frequently appearing groups in the martial arts world. In every impressive fierce battle, there are always Shaolin monks and their mysterious martial arts. In the world of routing, there are two major schools: IGP and BGP. Perhaps we can say that BGP should be a porter of routing, responsible for the transmission and exchange of routes between multiple autonomous systems. It can be said that "network routing comes from IGP". IGP is definitely a school with profound internal strength and many tricks. It uses complex tricks and internal strength to generate routes. It is the source of routing. It is further divided into: OSPF, IS-IS, RIP, IGRP, EIGRP and many other schools. The two major schools of IGP and BGP live in peace, each doing their own job, and have lived in peace in the world for decades. In recent years, SDN has been surging. Under the wave, traditional routing seems to be at the end of the road, or is it rejuvenated?

[[226188]]

Routing Overview

BGP's position and functional role in the network, combined with its extremely flexible scalability, has demonstrated tenacious vitality under the SDN wave. We will not discuss this for now, and here we will focus on IGP.

IGP routing is generally divided into two categories: link state routing and distance vector routing, which are essentially Dijkstra algorithm and Bellman-Ford algorithm. Here is a brief summary of the current status of IGP routing protocols:

• RIP is only suitable for small networks and there are few application cases today.
• IGRP is dead, and IOS has long since removed this feature.
• EIGRP is widely used in specific industries, such as the financial industry.
• HWMP, OLSR, etc. applicable to wireless MESH networks. Due to the application limitations of wireless MESH networks themselves and various alternative technologies, their routing applications are less common.
• Traditional link-state routing protocols are used on a large scale and are de facto mainstream protocols, such as OSPF and ISIS.

Traditional link-state-based routing protocols have many advantages, such as being applicable to large networks, having no hop limit, fast convergence, and being naturally loop-free. They have basically become the most widely used routing protocol category in the existing network. It is worth mentioning that OLSR (Optimized Link State Routing) applied in the wireless MESH field has made bold innovations and optimizations to the traditional OSPF and IS-IS link-state routing protocols, and has reduced the amount of message flooding through the MPR (MultiPoint Reply) mechanism. Under the scale of large-scale network topology, the optimization is extremely obvious, and it is an elegant routing protocol. OLSR relies on wireless MESH networking with fewer application scenarios, and the wireless MESH network itself has many alternative technologies, which makes it less used in the existing network. Especially with the rise of SDN, it is foreseeable that OLSR will decline before it rises.

The media once selected Dijkstra's shortest path algorithm as one of the ten basic algorithms that rule the world, which is not an exaggeration. Without this algorithm, the current Internet would not work.

The core of OSPF, ISIS and other routing protocols are essentially single-source shortest path algorithms. The synchronization of link state across the entire network and the collection of network-wide topology information have become the main differences in the protocols, which are reflected in the process of neighbor establishment and LSDB synchronization. This is also a behavior that must be regulated at the protocol level. Putting aside the differences in link state synchronization mechanisms, the core ideas are as follows:

Once the device neighbor relationship is established and the link status information is synchronized, the shortest path calculation is performed with the current device as the root to generate the routing table.

After the routing table is generated, the table entries are put into the hardware forwarding chip to generate a forwarding table to guide the hardware to forward at high speed. Traditional routing switching equipment is based on this idea, and the various data communication features derived from it are all developed around it.

We can see that the traditional routing principle is not complicated, but OSPF is undoubtedly still the most complex routing protocol to date. The OSPF protocol standard RFC 2328 document is nearly 300 pages long, which is enough to illustrate this point. As a result, an ecosystem has been derived around the routing field. Various certifications of traditional equipment vendors, such as CCIE, HCIE, H3CIE, etc., have become a popular profession. Getting these certifications, especially CCIE, means a high-paying job. Although CCIE is on the decline today, it is still hot. It can be predicted that under the wave of SDN, various network engineer certifications such as CCIE will gradually cool down. Simple routing principles have extremely complex realities. Functions and features emerge in an endless stream, bringing the vitality of traditional routing equipment. Is it a technical threshold deliberately created by traditional equipment vendors? So that only a few can play with it. Isn't the essence of CCIE and other certifications to drive the sale of routing equipment? Applications in various scenarios such as IGP + MPLS have hundreds of configuration commands. How can users tolerate it? The ecosystem has been established and customers cannot get rid of it. SDN is an opportunity, but will equipment vendors let it go easily?

Routing Change

Under the SDN trend, the network needs to be smarter, simpler, and more flexible. The heavy and complex functional characteristics of traditional routing equipment are difficult to cope with. Faced with the huge amount of traditional routing equipment in the existing network, SDN reshapes the entire network and faces the following problems:

1. How to transform traditional network routing into SDN?

2. How to smoothly transition between traditional routing and SDN network routing?

3. How to define the forwarding path by software?

4. How to solve the problem of insufficient backup of traditional routing paths?

Regarding question 1, I personally think that it is impossible to achieve true SDN transformation based on traditional routing equipment, but we still have to make it look a little like SDN to customers, and it looks like SDN on the surface. At present, the mainstream practice of equipment vendors is that the SDN controller centrally controls the configuration, and the traditional routing equipment is used as the forwarding plane. The SDN controller focuses more on the unified and centralized operation and maintenance level. The variety of equipment vendors, coupled with different understandings of SDN and different implementation methods, has led to the SDN southbound interface being all-encompassing, and in fact there is no unified standard. The real power of SDN lies in question 2, the connection and transition between traditional routing networks and SDN networks. There are various solutions, which will not be discussed in depth here. The simplest and most direct solution is that the traditional routing device is connected to the SDN forwarding plane. The SDN forwarding plane directly sends the routing message Packet-In to the SDN controller, and the SDN controller is handed over to the upper-layer APP for routing protocol processing, playing the role of docking with the routing device. The SDN controller sends the routing protocol message to the forwarding plane through Packet-Out and transfers it to the traditional routing device to complete the docking. The SDN open source controller ONOS has a similar solution for BGP networking as follows:

Based on questions 1 and 2, the forwarding plane still needs to run routing protocols at this stage, otherwise it will not be able to cope with the complex networking environment of existing network devices, and traditional equipment vendors will not cut off their own arms. In the ideal SDN, the control plane is completely shrunk, and there is still a long way to go, so we can only compromise with reality. Let's think about a less concerned issue: In a pure SDN network, how does the forwarding plane connect to the SDN controller and how is the control channel routing generated? This is a question of which came first, the chicken or the egg. From a rational point of view, a pure SDN network is not feasible and still relies on the traditional network.

Question 3: In an ideal SDN network, routing control is entirely in the SDN controller, so it is relatively easy to define the path using software. Combining traditional equipment and TE traffic engineering, Segment Routing is undoubtedly the best choice. For a detailed introduction to Segment Routing, please refer to the following article: Segment Routing will help SDN reshape new networks

Question 4: The traditional routing path backup feature has been developing. The shortest path LFA (Loop-Free Alternate) is a widely used technology. However, this technology cannot calculate the backup route in many cases. More and more complex backup route calculation methods have been derived, such as RLFA (Remote LFA), MRT (Maximally Redundant Trees) and other technologies. These new technologies waste more resources than route calculation in order to produce backup routes, which greatly reduces the convergence performance. Personally, I think these technologies are limited to the research level and will rarely be used even in the future production environment of the existing network. However, under the SDN architecture, the generation of backup routes will have unique advantages, which depends entirely on the space for the controller APP to play. In theory, all existing backup routes can be generated. In the SDN controller technology, combined with Segment Routing, the calculation of backup paths will be easy.

Outlook

Traditional routed IP networks can be widely deployed around the world thanks to a distributed protocol system. Even if one or more routers crash, the IP network can still continue to operate, which is amazing. However, it is this distributed system that makes it difficult to implement many special global policies that would have been easy. The "IQ" of IP networks is actually very low. What's more troublesome is that there is a gap between the application layer software and the router software in the network control, making it difficult to implement application policies fully automatically.

The introduction of SDN is trying to solve these problems. However, the mechanical and aggressive introduction of SDN may even bring more problems. In the wide area network composed of routers, one move affects the whole body, so the introduction strategy of SDN is relatively conservative. The conservative performance is to try not to touch the forwarding plane of the router, and try to maintain compatibility on the control plane. In a hybrid system operation mode, the control plane running on the router is gradually transferred to the centralized operation on the controller, and through the hierarchical controller and reasonable northbound interface, the demand for full software automation is realized.

SDN has passed the hype stage. Traditional equipment vendors, radical researchers, operators, etc. have gradually calmed down. Traditional routing, as the "authentic Shaolin", is still indispensable; but SDN can greatly reduce the complexity of network equipment, stop the trend of development towards complexity, and make the network simple, automated, and intelligent.