What is the difference between SDN network and traditional network?（2）

If you want to know previous content of this article. Pls check the link

What is the difference between SDN network and traditional network?（1）

MPLS based network

When the ATM protocol and IP network PK dropped ATM, the market failed due to many reasons such as the complexity of the ATM protocol and the high cost of the equipment. IP network also sincerely envied the advantages of virtual circuits realized by ATM. For connectionless IP, if it can form an on-demand virtual circuit, it can provide different quality of service and forwarding path according to user characteristics, which is very attractive. The power of things, coupled with some bottlenecks in forwarding performance at the time, combined factors contributed to the birth and popularity of MPLS.

When MPLS began to be planned for use and deployment, the two basic concepts represented what the operators had been waiting for for a long time.

FEC forwarding equivalent class

FEC (Forwarding equivalence class) provides the same forwarding processing method for a group of data packets of the same classification. The same classification mentioned here, the same destination address is only one of them, and free network forwarding based on different standards has always been one of the visions of network engineers.

LSP: Label-switched path

The LSP classifies the packets based on the FEC, and implements the end-to-end IP packet encapsulation. The unidirectional virtual circuit is forwarded based on the Label per hop. Just as the original intention of IP network to Frame Rely, ATM, SONET/SDH, MPLS Dividing the network into two parts, Core and Edge, the basic idea is that the Edge device encapsulates various requirements, and the Core part only performs label forwarding.

In the whole network, through the Label distribution mechanism, the Edge node replaces the destination IP address through the Label, and the forwarding plane abstraction is provided in the network.

The power of MPLS is the development of a range of services based on label switching.

VPN service: L3VPN, L2VPN, VPLS

TE: Traffic Engineering Services

Multicast service

The most successful of these is the deployment of a fairly extensive L3VPN service. The reason behind this may be the huge market demand for three-layer network isolation. Even in the absence of L3VPN Internet, IPsec, SSL VPN and other technologies have developed.

Other VPNs such as VPLS, although used in certain ways. However, due to its limited market demand and some shortcomings of MPLS network commonality and its own weaknesses, it has not been implemented on a large scale.

MPLS was ambitious in the past, with multiple WGs in the IETF, and numerous RFCs evolved at the same time. But the complexity is complicated, and in the end, a lot of content becomes a paper article on the high and the low. At least on the level of Chinese operators and enterprise networks that can be reached, it cannot be reached on a large scale.

Common weaknesses of MPLS services

The complexity of multi-layer protocol accumulation and the cooperation between protocols

For example, to implement a VPLS network, at least the following protocols are required:

IGP - the lowest PE/32 route

LDP - for outer label, PE addressing

BGP - For Internet services, although not required for VPLS, BGP for IP services is actually a standard for operators.

MP-BGP - for topology discovery and provision of inner tags

When multiple layers of protocols coexist, the vulnerability of network services comes from whether each protocol on each device works. At the same time, when there are problems with the cooperation between different protocols, it may also cause problems, such as the synchronization problem between classic LDP and IGP.

In addition, VPLS itself has some weaknesses, such as:

Need Full-Mesh

Implement MAC address learning on the forwarding plane.

Multihoming that does not support CE

MPLS-TE

The problem that TE most hopes to solve is that the traffic caused by the same shortest path view of the IGP routing protocol is too concentrated on a small number of paths, and the utilization of the entire network needs to be improved. Therefore TE needs to establish and maintain a large number of end-to-end tunnels. The cost of maintaining these tunnels is a lot of money. In the distributed network architecture, TE has not solved the problem of tunnel calculation/maintenance/disordering. TE is also limited to use in some carrier networks.

Of course, after many years, we also know that the ideal FEC is just the beauty of imagination. In reality, 99.9% of the messages are still forwarded according to the destination address. Limited by chip, equipment, protocol / standardization, equipment, operation and many other factors.

Through a simple analysis of the MPLS service, we can easily see that MPLS uses Label to better solve the abstraction of the forwarding plane, but does not solve the control plane abstraction. It is the complexity of its control plane that affects the network and weakens its scope of deployment. The complexity of the control plane is the essence of it, or each section of each protocol, each satisfying a demand. To do VPN, you need L3VPN or VPLS, you need to do traffic engineering, you need TE, you need QoS, you need IP or MPLS Qos, and so on. If this is all, is your device supported at the same time? Do you dare to deploy? Are you willing to operate this network? I believe that most of the network workers are faced with these problems in the face of operators, integrators and manufacturers.

So many technologies are still called Advanced Technology, not universal applications, many years later. Many times, in general, can you implement a function? Yes, but there are too many preconditions: network design considerations, complexity of a large number of configurations, hardware performance support of devices, configuration methods and default behaviors of different devices for different features, interoperability and compatibility. The operability of the barrier.

Over time, it can become a theoretical, idealized statement. It’s not the ability to simply land.

Is Segment Routing a white knight?

In the past two years, SR has emerged, making people look bright. There are some traditional network saviors. SR has two significant advantages over the previous phase of MPLS-related protocols.

It is indeed very delicate to use the MPLS and IGP protocols, greatly omitting the need for label distribution protocols. Simplified protocol and network design.

The SR is not responsible for the calculation of the path and is only responsible for forwarding. The middle node is really based on Label forwarding only, by embedding the path map in the packet header on the multi-layer stack of the headend device. It's like starting with a few tips. Every time you go to a key node, open a kit and see the next key node name and path. SR therefore goes beyond the shortest path algorithm to limit the forwarding of messages without the need to maintain a cumbersome number of tunnels. SR is more like a master of navigation, as long as there is a clear route (path) to sail as required.

But it is clear that although SR is good, someone still needs to calculate and design the way forward. Who is more suitable to do? Obviously it is SDN.

SDN network

The decoupling of the control plane and the specific network equipment is the most important feature of SDN. The control plane can only form a powerful brain after decoupling. After adapting and orchestrating the ever-changing business layer requirements, it can be forwarded into devices in different ways. Undoubtedly, whether the control plane is concentrated, how to maintain the HA after the concentration and the rapid convergence of the topology changes of the whole network are all problems.

From this perspective, the concentration of the control plane is not so much the characteristics of SDN, but rather the inevitable and necessary cost of SDN at this stage.

From a certain point of view: the content that the route needs to complete is mainly the following three items:

Establish a topology, equivalent to a network map

Passing routing information of different Nodes, equivalent to residents of different villages

According to different needs, the calculated path from A to B

From these three points, SDN has an advantage over routing protocols. Because there is a natural view of the whole network, because the information of each Node is known naturally. Because of the docking business orchestration, it is possible to integrate various requirements and perform global routing calculations. However, it is not completely the case from the landing situation. It is mainly the first point. When the topology changes, the convergence ability of SDN is not as fast and effective as the traditional routing protocol.

Web service vs. network for you

By simply combing through the three development stages of the network, we can clearly see that the network changes from the IP network that originally provided accessibility to the MPLS network with basic network segmentation services, to the demand-centric, on-demand Provide a service SDN network.

It turns out that creating separate protocols for each need does not work, increasing the complexity of the control plane and the ability to evolve rapidly.

The network has been criticized and complained about by applications, the speed of deployment and the flexibility to meet multiple needs at the same time. The larger the network, the closer it is to the simple, basic reachable service, not the instant customization service that can be provided for an application, a temporary connection, or an important user. The cost of providing such a service is not scalable at the design level, it is hand-built at the operational level, and it is catastrophic at the troubleshooting level.

How to make the network work for the upper layer in more dimensions and finer granularity. In addition to having a unified northbound interface, it is possible to use a brain to understand the topology, manage the equipment, and design the path is the possible basis for everything.

Comparative analysis of MPLS and SDN networks from the perspective of network life cycle

The Cisco Service Department has a definition of the network life cycle. The initials are PPDIOO, which stands for P (Pparate), P (Plan), D (Design), I (Implement), O (Operate), and O (Optimize). The first two steps, Prepare and Plan, start with the requirements, and then clearly define the strategy, resources, mission-critical, conceptual design and milestones. From a technical point of view, it can be merged into the design.

The ITIL service cycle definition is divided into five phases, namely

Service Strategy Strategy

Service Design design

Service Transition

Service Operation

Continual Service Improvement Continuous Improvement

Also from a specific technical level, strategy and design can be combined into a design.

The usual network practice experience can also be summarized as design (including testing) - deployment - operation and maintenance (including troubleshooting) - optimization of four phases.

design phase

The final product of the design phase of the MPLS network is LLD (Low Level Design).

Usually, in HLD (High Level Design), the basic layout, traffic, interworking, QoS, etc. are required to be quantified in the LLD and the landing plan is given.

The design of the MPLS network is simply the routing protocol + device = configuration. Because you need to support multiple protocols, you need to consider more.

Equipment level:

To test and verify protocol compatibility for devices from different vendors

Consider the performance of the device after supporting multiple protocols

Protocol implementation level:

In the coexistence of multiple protocols, there is a conflict between the individual designs. After MPLS, the default impact of different vendors on the QoS model is different, and manual configuration must be consistent.

The SDN network only needs to consider the deployment location of the controller, the number of ports of the network element device, and some necessary supporting services. Simplified network design.

Deployment phase

Deployment usually includes:

Allocate and manage various virtual resources:

IP address segment, VLAN, AS No., etc

Configuration template:

After the configuration file is divided into blocks, the configuration contents are basically the same, or after the initial configuration, the devices are individually logged in and configured.

Even if there is a configuration script, the MPLS-base network needs a complex check process based on the protocol status to determine that the network deployment is correct.

In contrast, SDN network equipment is simple to configure, and can usually be easily automated. Some devices can also be automatically deployed after startup. The essence is also realized by the many-to-one relationship between the network element device and the controller.

One of the hard things is the design and deployment time of the MPLS network, taking things into consideration, testing and doing it, writing and writing documents, 1-2 months is really not slow. Plus the equipment arrived, the installation, no more than 3 months can not be built. The network has been built for too long, which is something that everyone has experience and is helpless. From the root, I understand the problem or the coupling between the protocol/configuration/device is too deep.

Operation and maintenance stage

Monitoring, changing and troubleshooting are the three tasks of the operation and maintenance phase.

monitor

From the commonality, topology discovery and change, link delay, packet loss, traffic monitoring, analysis, and alarm. It is an indispensable feature of both types of networks.

In terms of difference, MPLS network operation and maintenance needs to monitor the status of multiple routing protocols. Because it is fully distributed, each protocol may have Peer on each interface. It needs to monitor multiple states or remove faults afterwards. The underlying faults have an impact on the upper layer protocol. How to manage and block a large number of alarm events is a challenge.

SDN has a little advantage in that it can perform more white-box monitoring, that is, to monitor the operating status of the system by monitoring the internal performance indicators of the system. Because from the south, SDN only needs to monitor a few protocols, and the monitoring is relatively simple. In the face of business changes, the API can be satisfied anytime, anywhere. The main complexity is concentrated on the control plane and service orchestration. The monitoring is also mainly focused on the control plane robustness, user service status, and consistency of control and forwarding. A large number of path calculations and re-optimizations due to underlying link failures in large networks will require controlled timely response. Web interfaces for end users will in turn require real-time response and analysis of various request and configuration changes. To a greater extent, the network is a complex computing and business system.

The advanced stage of monitoring is to associate and link the underlying data, and the second is to mine user data. From this perspective, both networks still have a long way to go.

change

If a large number of long-distance faults come from the underlying link, which is caused by the bulldozer smashing the fiber, the artificial configuration error caused by the change is another major source of the fault. There are many reasons for the change (software upgrade, hardware replacement upgrade, business deployment, structural optimization, etc.), and finally will be implemented to adjust traffic and change configuration.

After the network deployment, as the operation and maintenance time elapses and various service requirements meet different requirements on the network, it is difficult to maintain a unified configuration template. Various temporary and non-standard configuration requirements are gradually taking root and sprouting on different devices. In the end, no one dared to delete temporary configurations that might not have been used. Those configurations are also semi-permanent on different devices. The traditional network configuration is long on the device, and the ins and outs of various configurations are only clear to the parties. Over time, changes in people, equipment, and demand can cause the configuration to be out of line with the actual situation, and it is out of touch with the existing network operation and maintenance personnel. Just talk about which network does not have a bunch of unclear policies, a bunch of ACLs that cannot be deleted.

There is a saying that when netconf+Yang is used, the traditional CLI is replaced and can be automated. However, the actual situation is that regardless of the promotion and landing time of Yang, even if all the vendors' devices can be delivered through the Netconf interface, the consistency of complex configurations, configuration check and post-check configuration delete rollback and configuration management are still challenging. task. Therefore, the automation of the configuration is not = SDN, the bottom line is the problem of excessive distributed complexity and tight coupling of physical devices.

SDN basically got rid of the problem of device configuration. Infrastructure data can be implemented entirely in the GUI through self-discovery and initial definition. The service data is implemented through the GUI and the API. When the software is upgraded, the front end of the control plane, the back end, the service orchestration, and the components of the underlying controller can be upgraded separately or uniformly. No significant impact on forwarding.

Troubleshoot

MPLS network troubleshooting fully reflects the complexity of the network, understanding, application and analysis capabilities of protocols, vendor equipment, specific configurations, network topology, and troubleshooting tools. The troubleshooting time has a pressure on the operation and maintenance and production network business. This, in turn, affects network design and may tend to be more conservative in selecting technologies and devices. Expect to reduce the difficulty of troubleshooting.

SDN troubleshooting needs to be combined with Devops and solved by software. Complexity will never disappear, it will only shift. One of the challenges faced by traditional network workers is that the original troubleshooting tools and ideas on the CLI may not be applicable to SDN, API or other software tools. Need to hold an open mind and gradually explore in the process of operation and maintenance.

optimization

Network optimization is not inevitable. Sometimes, the subtext of optimization is that the existing network cannot meet the new requirements, and it is necessary to update the equipment to achieve the new feature to meet the new requirements.

When the MPLS Base network is optimized, the troubles of the previous design, deployment, and operation and maintenance have to be repeated.

The SDN network is basically two points:

1. Scale-out of the controller cluster expands horizontally with the expansion of the network scale

2. Scale-out horizontal expansion of NE devices

In summary, the operation of the network wants to be good (low failure rate, stable, fast speed, high quality). Before the SDN technology is produced, it seems that a network team belonging to the elite club is required to complete the content.

Complex design and complete equipment testing

Powerful operation and maintenance, 7/24 Ready to handle complex faults at any time

Powerful tool support

Strict and complex change system and professional team

Due to the size of the network, operators can get the highest priority support from vendors or integrators, and have a fairly complete team of engineers focused on network operation and maintenance. This is why the most powerful operators in the past years have always been the leader in network operation and maintenance. But conversely, this is also evidence that the network needs to improve. The network needs to be simpler, more flexible, faster to deploy and adapt to the business, more convenient and white-boxed monitoring performance.

The SDN network basically solves the path calculation according to the business required in the software framework, and the automation of the southbound interface. Of course, as a software concept, SDN will inevitably face more software problems directly affecting the network. Deployment and operation and maintenance become more software deployment and operation and maintenance. All the software will encounter headaches, SDN will encounter.

Conclusion

In fact, there is a big problem in the MPLS network that has been felt for so many years. The routing protocol (routing strategy), the deep coupling between devices and configurations.

Routing protocols (routing strategies) are implemented through local configuration

The implementation of the configuration on different vendors' devices is different.

Based on local configuration, complex content, multi-protocol and other reasons, automation and programming are difficult and incomplete.

In the end, SDN should be implemented and landed. In the past two years, many practical explorations have been started. The big vision has already been made, but there are still many disputes about the specific landing and approach. Is it continuing to develop into SDN along the MPLS network, or is it more thorough on the essential issues. Still waiting for time to test. Data networks have become the wind power and hydropower in the data age. The network and the different technologies of different operators and different scales have different development technologies and applicable technologies. The generation and occupation market of each generation network meets the specific conditions at that time. logic. Perhaps the pure IP network, MPLS-Based network and SDN will coexist for a long time and develop together.

"The future has come, but the distribution is not so average." We can believe that things happening in the cloud will definitely happen on the network. On-demand, dynamic, resource pooling, API, highly automated and flexible business abstraction are sure to become the main features of the network.

Know more useful tips or study guide related to IT certification exam and Network technology from SPOTO now.