Shulou(Shulou.com)06/02 Report--

As more and more enterprises embark on the journey of business going to sea, the urgent problem to be solved is: how to achieve the rapid mobilization of IT resources and the smooth flow of information around the world? If you choose long-distance transmission over public network, the experience of network delay and jitter is very poor; direct connect is expensive, and multiple direct connect will cause difficulties in operation and maintenance; with the popularity of multi-cloud deployment, cross-cloud transmission is also one of the difficulties that enterprises have to face.

The enterprise goes out to sea, the network comes first. Under this background, UCloud's SD-WAN product Rome Rome arises at the historic moment. Compared with the traditional scheme, Rome has great advantages in cost, performance, availability and so on through the underlying architecture design. For example, the use of Segment Routing to improve forwarding efficiency, intelligent link scheduling to achieve second routing, and so on, the following will focus on.

-01-

Rome: SD-WAN-based Global Network acceleration

SD-WAN technology is a specific application of SDN under the wide area network. By eliminating the traffic management burden of physical devices and transferring it to software, it can obtain the flexibility of deployment and management, and help users minimize the trouble of managing infrastructure and connections.

| | Rome network composition: core network + access network |

(1) Rome core network

Based on SD-WAN technology, Rome core network relies on UCloud global data center and cross-regional and cross-operator networks to provide users with global access to the nearest and link dynamic scheduling, to achieve end-to-end highly stable connection, to avoid backbone network failures, cross-regional / cross-operator congestion and other problems such as slow response, packet loss and so on.

Figure: distributed global Rome core network

(2) Rome access network

Rome provides access to a variety of scenarios, including Direct Connect, multi-cloud, hybrid cloud, Internet and so on. Users need to choose the most appropriate access method. Rome connects users to the forwarding nodes of the core network in the same region or adjacent regions, and the Rome core network distributes the traffic to the destination region. Users can choose the following access methods:

When the initial traffic of ① is small, you can choose EIP access, which belongs to the interconnection of EIP in the same region and will not affect the network quality; when the traffic of ② is large, you can choose direct connect, which can guarantee the network quality to a great extent, and when the access direct connect fails, Rome supports seamless switching to EIP to ensure the normal business of users. Direct Connect can be connected with one button through the UCloud console, and the subsequent Rome will also launch the EIP button with one button, so that users do not have to wait for a long construction period.

| | Control plane is separated from data forwarding plane |

Because of the particularity of long-distance transmission, SD-WAN is divided into control plane and data forwarding plane according to its function. This separation can greatly increase the agility of network services and can migrate more intelligently from the data plane to a more abstract and programmable control plane. Rome is also composed of control plane and forwarding plane to ensure high and reliable data transmission.

(1) Rome control plane design

The control plane is mainly responsible for the exchange of network signals and the selection of packet routing, it also includes the system configuration and management of the equipment, and is also responsible for ensuring the smooth control pipeline of the business surface.

Figure: UK8S dual-live architecture in the same city

As shown in the figure above, the Rome control layer adopts UK8S dual-live architecture. Multiple replicas of each application are peer-to-peer deployed in UK8S clusters in two data centers. Domain name resolution service DNS is used to channel ingress traffic into two clusters, effectively preventing and reducing business unavailability caused by a single data center egress or single replica failure, and ensuring business stability and high availability at the cluster (data center) and application level.

In addition, the control layer consists of API, TASK, CONFIG, Monitor, DB and Analysis. The execution logic of each component is shown below: when the API receives a data forwarding request, it will be sent to each UK8S cluster. TASK is the main logic executor, CONFIG is responsible for the configuration and distribution of each network element on the forwarding side, DB Agent is responsible for storing the data that needs to be persisted, Analysis schedules the Rome core network to ensure that user traffic can Run on the optimal path, and finally Monitor completes the monitoring of the status and quality of each service and Rome core network.

Figure: schematic diagram of Rome control plane execution logic

Among them, Analysis and Monitor play a key role in link scheduling, which will be described in detail below.

(2) graphic design of Rome forwarding

The forwarding plane mainly carries the forwarding applications and user data of the network, and is composed of high throughput forwarding nodes. Rome implements packet forwarding nodes with high throughput and low delay based on DPDK, and a dedicated line is used as the main line between any two nodes to ensure the high reliability of the line network.

Figure: Rome forwarding surface architecture

As shown in the figure above, each forwarding node and access node are composed of two nodes and are active and standby to each other. When the primary node is unreachable, it automatically switches to the standby node, which can switch in seconds, thus ensuring the network quality to the maximum extent. In addition, a ring network is formed among the regions of the Rome, and the path is uniformly scheduled by the Analysis according to the path quality.

It is worth mentioning that Rome also uses Segment Routing (hereinafter referred to as SR) technology.

SR is a new type of MPLS technology, in which the control plane is implemented based on IGP routing protocol extension, and the forwarding layer is realized based on MPLS forwarding network. The advantage of SR lies in simplifying MPLS control protocol, reducing resource occupation, simplifying network operation and management, and enhancing path adjustment and control ability.

Rome forwarding plane uses SR routing to achieve traffic forwarding. The control plane calculates the optimal path between two points according to the real-time network quality, and assigns labels to adjacent nodes (used to identify routing adjacency links in SR networks, which is the main label type used by SR-TE, and link labels have certain directionality, which is used to guide packet forwarding, which is only valid locally in the source node), and assigns label stacks to the whole path (a set of label sorting, used to represent a complete LSP). The forwarding node identifies the label carried by the message and routes according to the label. The application of SR in Rome greatly improves the forwarding efficiency of the network, reduces the delay, and ensures the user experience to the maximum extent.

-02-

All roads lead to Rome: intelligent Link scheduling under Real-time Monitoring

In addition to high stability and high performance, the flexibility and agility of the network are sometimes more critical: how to quickly achieve the optimal link scheduling according to the needs of users and the real-time state of the network, so as to achieve the seamless flow of data around the world? The answer of ☟ Rome is the intelligent link scheduling technology in routing and handover. For example, when users are using two adjacent nodes of the network with a sudden increase in delay, Rome will switch traffic to adjacent nodes with smaller delay in seconds, thus ensuring the overall network quality.

| | create line quality metadata |

First of all, Rome realizes a good foundation for real-time data monitoring by establishing fine-grained line quality metadata, and also helps network operators to grasp the dynamics and flow of global business data more easily and in real time.

Figure: real-time monitoring of line quality metadata

As shown in the figure above, Rome is divided into access node, forwarding node and DB to complete the establishment and real-time monitoring of line quality metadata. The access node & forwarding node is responsible for regularly reporting line packet loss rate, line delay, bandwidth load, access point CPU utilization, access point disk utilization and other data. DB stores all data for background real-time analysis and monitoring.

Each access node and forwarding node have active and standby nodes, and the access node and the forwarding node are mesh connected, and the forwarding nodes are also mesh connected. As shown in the figure below, all forwarding nodes form the Rome core network.

Figure: Rome core network

| | Intelligent routing |

Based on the dedicated lines distributed around the world, Rome provides more flexible and reliable routing through link intelligent detection technology. Intelligent routing is divided into two aspects: access node routing and forwarding node routing.

(1) routing of access nodes

Take region 0 as an example, as shown in the figure below, user VPC-- > access point VPC-- > forwarding point VPC is connected to the master node by default, and TC rules are created for the user VPC at the access node. The access node and the forwarding node encapsulate the message through a tunnel.

Figure: user VPC-- > access point VPC-- > forwarding point VPC to get through

Then, Rome will carry out link detection in the Local access node and Remote access node respectively, persist the link detection result, pull the link detection result by Monitor, and select the path between the access node and the forwarding node according to the result.

The link detection results include the following dimensions: accessibility, packet loss rate and delay from Local access node to Local forwarding node; reachability, packet loss rate and delay from Local access node to Remote access node; load of Local forwarding node. In the end, there are three types of situations:

① if the Local primary access node is unreachable to the Local primary forwarding node or the network quality is poor, the path from the Local standby access node to the Local standby forwarding node is selected; in ②, if the Local primary access node is unreachable to the Remote primary access node or the network quality is poor, then the Local standby access node is selected to the Remote standby access node; if the load of the Local primary forwarding node is high, the Local backup access point node is selected to the Local standby forwarding node.

(2) routing of forwarding nodes

Taking region 0-> region 2 as an example, the available paths of region 0-> region 2 are region 0-> region 1-> region 2 and region 0-> region 3-> region 2. The master / slave forwarding nodes of each two adjacent regions are interconnected, that is, there are 4 link of the two adjacent regions.

Figure: region 0-> region 1-> region 2 forwarding

Rome also detects the link of the adjacent forwarding node, persists the link detection result, and the Analysis pulls the link detection result, and selects the link of the adjacent forwarding node according to the result.

The link detection results include the following dimensions: reachability, delay and packet loss rate between adjacent forwarding nodes; forwarding node load; tariff. The steps of link detection and scheduling are as follows:

① calculates the available paths of source forwarding node-> destination forwarding node based on the information of source and destination forwarding nodes; ② calculates a cost value based on the delay, packet loss rate, load and tariff of each link of all available paths, thus calculating a total cost value for each available path ③ selects the best path according to the cost value of each available path, assigns a label to the forwarding node through which the whole path passes, and selects the next hop according to the label when the traffic enters a forwarding node; ④ cost calculation principle (the smaller the cost, the higher the path priority): the cost is maximum when the ► network is unreachable; when the ► network is reachable, the direct connect cost=a delay + b packet loss rate + c forwarding node load EIP cost=x (a delay + b packet loss rate + c * forwarding node load) when the ► network is reachable, x > 1 due to tariff reasons.

| | Intelligent switching |

Intelligent handoff is divided into access node handover and forwarding node handover, and all handover actions are completed in seconds, when the network failure in use, it can be guaranteed to switch to a normal line at the first time. When all direct connect of link fails, the traffic will be switched to EIP in time to provide users with stable and reliable network quality.

(1) Handoff of access node

Monitor calculates the optimal path according to the routing principle of the access node. When the calculated optimal path is different from the current path, compare the cost values of the two paths:

① (current path cost value-optimal path cost value) / current path cost value

Threshold, the user VPC route is directed to the standby access node, and the roles of the active and standby access nodes are exchanged.

(2) Handoff of forwarding node

After each route selection, Analysis makes a slice of the path cost value, and selects the optimal link of the current line (the effective path of two adjacent forwarding nodes) according to the routing principle of forwarding nodes. When the optimal link is different from the current link, compare the cost values of the two link:

① (current link cost value-optimal link cost value) / current link cost value

At the threshold, the optimal link is assigned a label, and the optimal path information is updated synchronously, and the traffic is introduced into the newly calculated optimal path.

-03-

Rome application example: overseas cross-regional multi-cloud e-commerce scenario

Currently, Rome supports application deployment in multiple scenarios, including multi-cloud scenarios, cross-cloud disaster recovery, and so on. Here we take overseas cross-region multi-cloud scenarios as an example.

In order to ensure the high reliability of global business, a cross-regional multi-cloud e-commerce business chose a multi-cloud solution to support global business deployment at the initial stage of network construction: the warehousing system is distributed in Tokyo area of a cloud manufacturer, and the Web business is distributed in Oregon region of b cloud manufacturer. However, due to the increase in the volume of business in the later stage, Web business needs real-time access to the warehousing system, which requires two networks distributed in different cloud manufacturers to achieve fast and reliable communication around the world.

There are differences in interface protocols between different cloud vendors, so it is difficult to get through directly, let alone to achieve the purpose of long-distance real-time transmission, which should be an impossible demand. In the end, the e-commerce has successfully adopted the Rome solution, and the network architecture scheme is shown in the following figure.

Figure: overseas cross-regional multi-cloud e-commerce application example

At the same time, the use of Rome is also very simple and convenient. Even in such a complex cross-region multi-cloud scenario, users only need to do some simple operations on the console and do not need to care about routing configuration and other issues to achieve one-click access to cross-region multi-cloud networks. After the user operation, Rome will assign the access node to the user in the region that needs to be connected, and connect the access node to the nearest forwarding node in the Rome core network, so as to achieve long-distance interconnection under cross-regional and cloudy conditions.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.