Alibaba Cloud researcher Zhu Shunmin: A cloud-native network born from the cloud

Note: The author of this article Zhu Shunmin (named Jianghe), a researcher of Alibaba Cloud Intelligence Business Group and head of network products, has many years of technology research and development and management experience in the fields of software-defined networking (SDN) and network function virtualization (NFV). Leading the team from scratch, creating the industry's most abundant cloud network product line, and developing the cloud network platform of Alibaba Cloud Feitian System-Luoshen, which has undergone the test of Double Eleven for 10 years.

The widespread use of cloud computing is changing the definition of cloud native. Cloud native is changing from a narrow definition of cloud native, namely "microservices + containers + continuous delivery + DevOps" to a broader definition of cloud native, that is, "cloud-born Software, hardware, and architecture are truly cloud-native". As the standard IaaS core product line of cloud computing, the cloud network was born from the cloud on the first day, and it was born as a cloud. It is a truly cloud-native network.

meets the cloud computing of

In 1969, the Advanced Research Projects Administration of the US Department of Defense established a network called ARPAnet, which is generally regarded as the beginning of the network. The TCP/IP architecture and protocol formed by ARPAnet are still the foundation of the network world today. In the following decades, the network developed rapidly. By 2009, the mobile network had developed from 1G and 2G to 3G, and the applications we used had evolved from the earliest phone calls and text messages to graphics, chat, and shopping. The development of these applications is inseparable from the network, it can be said that the development of the network has driven the transformation of applications. In the same year, the Ministry of Industry and Information Technology officially issued third-generation mobile communication (3G) licenses for the three major operators. From this moment, China officially entered the 3G era.

In 2009, most of the network cards of data center servers were still 1000M. On May 10, 2010, Alibaba Cloud released the first commercial product-cloud server ECS, which officially provided public cloud services, marking China's cloud computing public cloud services Opening.

At that time, the Internet had been in development for 40 years, and it had entered middle age, and cloud computing was just born, and it was still a newborn baby. What kind of sparks will the middle-aged network encounter when it meets the new cloud computing? At that time, many people may not have thought that cloud computing would have such a big impact on the network.

At the end of 2012, with the development of virtualization technology, the virtual ratio of a single physical machine was gradually increasing, and very high requirements were placed on network equipment. At that time, there were no network equipment that could meet the business virtual ratio in the world. In addition, the network architecture at that time still had problems in cost, stability, performance and other aspects, which greatly affected the development of cloud services. At the same time, as more users go to the cloud, especially large Internet companies and traditional enterprises, the demand for network management on the cloud is also increasing. For example, when users deploy services in multiple regions on the cloud, they need to communicate with intranets in multiple regions, or users need to interconnect IDC under the cloud and networks on the cloud to build a hybrid cloud, etc. These network management capabilities are the traditional network architecture on the cloud at that time It's hard to satisfy.

There is a fierce conflict between the middle-aged network and the newborn cloud computing. The middle-aged network must innovate to meet the rapid development needs of the newborn "cloud computing".

In 2012, Alibaba Cloud took the lead in abandoning the traditional large two-tier traditional network architecture and designing a new network virtualization (Overlay) technical solution. In 2014, Alibaba Cloud's virtual private network VPC product was officially launched. VPC is a product of the combination of middle-aged networks and new cloud computing. VPC is also a symbolic product of cloud networks, which verifies the establishment of the "network + cloud computing = cloud network" formula. Therefore, from the first day, the cloud network was born from the cloud.

Cloud network is a network with inherent cloud characteristics

Cloud networks are born from clouds and have the characteristics of clouds by nature.

first of all resource sharing . By virtualizing network resources, using Overlay technology to achieve multi-tenant sharing of network resources, but the tenants are safely isolated. Overlay technology allows each user to use the cloud network like a virtual machine, including the use of virtual switches/virtual routers/virtual load balancing and so on.

followed by elastic expansion . For traditional equipment, the processing performance is determined when the equipment is bought back, and it is difficult to elastically scale. In order to cope with elastic scaling, the cloud network separates the control plane from the forwarding plane, and adopts the cluster method to support expansion, combined with network virtualization technology to provide each user with elastic network capabilities.

again a self-service . This is the biggest difference between cloud networks and traditional networks from the perspective of users. Traditional network requires professional network administrators to type CLI (Command Line) command line to configure, while cloud network API and console, users can easily complete cloud network construction.

finally pay on demand . Enterprises using traditional network equipment, even if the utilization rate is less than 30%, can only be over-purchased for a certain degree of flexibility. However, the cloud network adopts pay-as-you-go, and enterprises charge according to the actual usage. This is also traditional What the network does not have.

The development history of cloud network

The development of cloud network has gone through three stages. The cloud network 1.0 in 2009 is the cloud data center network; the cloud network 2.0 from 2016 to 2020 is the cloud wide area network, and the cloud network 3.0 starting in 2020 is the application-cloud edge One network. At different stages, cloud networks provide users with different product capabilities to meet users' network needs.

Cloud Network 1.0 is a cloud data center network , which provides users with the ability to build data center networks on the cloud. Users can create VPCs, virtual switches, and virtual routers with a few clicks of the mouse to build a data center network environment, and then realize Internet access through EIP, load balancing SLB, NAT gateway and other products.

Cloud Network 2.0 is the cloud wide area network . After users deploy business systems in multiple regions on the cloud, their natural demand is multi-region intranet interoperability. In addition, users also need to interoperate between IDC under the cloud and VPC on the cloud. The cloud wide area network can well meet the needs of users in these two business scenarios. In 2017, Alibaba Cloud took the lead in the industry to release the cloud enterprise network CEN, and users can build a global network very simply and quickly. In 2018, Alibaba Cloud also released a cloud-native SD-WAN product-Smart Access Gateway SAG, to facilitate users to quickly go to the cloud. Cloud Network 2.0 allows users to easily build a global network integrating the cloud on the cloud and off the cloud.

2020, with the development of 5G/IoT/edge computing technology, society will enter the era of Internet of Everything and Industrial Internet, cloud computing has become the infrastructure of the entire digital society , cloud network also opened the 3.0 era. cloud network, as the highway of the digital society, will expand to the edge and produce better synergy with applications, so that everything will be connected to the Internet and help the development of the industrial Internet.

Based on cloud-native Luoshen Cloud network technology

The cloud network was born from the cloud and built on the cloud. On the first day, it adopted a fully self-developed approach and embarked on a road based on cloud-native unique cloud network technology. Including large-scale SDN control technology, Sailfish software and hardware integrated forwarding technology, CyberStar flexible network element technology, etc.

self-developed large-scale SDN controller, so that users no longer worry about scale and complexity

The SDN controller is the key to scale and management efficiency. Under the cloud native trend, a large number of users use Docker and K8s based on the cloud, which poses a big challenge to the scale and density of cloud networks. The SDN controller can meet the super-large-scale demand of users for accommodating 1 million instances per VPC, and can also meet the user's demand for high-density flexible network cards for a single ECS. In addition, in the large-scale context of public cloud, management efficiency is also very critical. Including the real-time and ease of use of user configuration changes, these are the advantages of the SDN controller based on cloud native construction, so that users no longer have to worry about scale and complexity.

Self-developed Sailfish forwarding platform with integrated software and hardware, so that users no longer worry about performance problems

The Sailfish hardware forwarding platform is based on high-performance forwarding components such as dedicated programmable chips and smart network cards. Through the design of an integrated software and hardware architecture, the high-performance forwarding capabilities of the host and basic gateway are built. It includes core modules such as high-performance gateway XGW, MOC card, ALI-LB, etc., which are integrated in software and hardware, and are currently widely used in various scenarios of cloud networks. As shown below:

The XGW gateway is responsible for the aggregation and distribution of public network, private line, and cross-region traffic. The MOC card is the core component of the server's internal network and is responsible for the forwarding and exchange of ECS traffic within the server. XGW gateway and MOC card together build a virtual private network for customers. In this virtual private network, the typical scenario is as follows:

1) Users access Alibaba Cloud via the internet (public network), and typical products used are EIP and shared bandwidth;

2) The user IDC accesses Alibaba Cloud, and the typical products used are high-speed channels (dedicated lines);

3) Cross-regional communication on the cloud, such as the Beijing region ECS visits the Shenzhen region ECS, typical products are CEN;

The evolution of network equipment has always been in a spiral development trend, around flexibility and high performance, software and hardware are merging and promoting each other. From the initial hardware forwarding equipment, to the software forwarding equipment, to the current hardware offloading software and hardware integration forwarding equipment and the popular programmable forwarding equipment. Flexibility and high performance are like an invisible hand, guiding the continuous development of technology.

In this regard, cloud computing service providers are more driven, prompting the research and development of cloud networks to take the lead in the industry. Take Alibaba Cloud's self-developed high-performance gateway XGW with integrated software and hardware as an example. Some customers propose to build a hybrid cloud through high-speed channels (dedicated lines), and the demand for dedicated line bandwidth is 10Tbps! If you use the original x86 server to build, even if you use high-performance servers, considering a certain degree of redundancy, you will need nearly a hundred servers. Both the cost and the operating efficiency are acceptable to the service, which prompts the cloud network to develop software and hardware integration The high-performance gateway XGW.

Alibaba Cloud's self-developed software and hardware integrated high-performance gateway XGW

At first glance, XGW looks like a 2U server. The main components are the P4 programmable switch chip responsible for large traffic forwarding, the self-developed network card chip responsible for forwarding large entries, and the CPU module responsible for business orchestration and intelligent scheduling. At present, a single machine can support 3.2Tbps, and it has been used on a large scale online, so that users don't have to worry about performance issues.

The CyberStar flexible network element platform allows users to be truly flexible on demand and makes investment more effective.

As more large-scale enterprise users go to the cloud, users have higher and higher requirements for flexibility, especially in business scenarios such as live video broadcasts and e-commerce promotions, which often raise demand for flexibility that has increased by a hundred or even a thousand times. In addition, users have high expectations for the delivery time of new features, and the platform also hopes to meet user needs as soon as possible. In this context, the original technical architecture based on x86 servers cannot meet business needs. In the e-commerce big promotion scenario, there are many public cloud users, and the platform does not know how many users will make the big promotion. Once the user's flexible demand arises, a large number of load balancing SLBs need to be used. At this time, we will purchase and deploy x86 servers to expand the SLB cluster. It's too late.

Therefore, building elasticity based on cloud-native ECS is a natural choice. The biggest feature of CyberStar's elastic network element platform is that network elements are no longer directly deployed in bare metal servers or dedicated computing resource pools, but are deployed based on ECS on the public cloud. Building on ECS means "unlimited" resources and "unlimited elasticity". Business network elements no longer rely on traditional x86 physical server deployment, and solve the problems of long capacity expansion and difficulty in flexible expansion using physical server deployment. At present, the CyberStar flexible network element platform is widely used in core products such as load balancing ALB, NAT gateway, PrivateLink, CEN-TR, Global Acceleration GA, VPN gateway, etc. Application load balancing ALB provides 1 million QPS capacity per instance, and can be "unlimited" flexibility, truly allowing users to be flexible on demand.

Future Outlook

In 2020, IDC FutureScape believes that cloud-specific equipment is a major trend. IDC predicts that by 2024, more than 25% of new workloads on public clouds will use service providers' dedicated infrastructure components. However, in the field of cloud networks, this trend is faster. Cloud-native dedicated devices developed with integrated software and hardware technologies have become the mainstream of cloud network business scenarios. The CyberStar elastic network element platform built using cloud-native ECS is also expanding. Scale use.

In the future, cloud network technology will continue to evolve in these two directions. Relying on cloud native construction technology architecture is the core advantage of cloud network. cloud network, born on the cloud, born as a cloud, is a true cloud-native network. (end)

Description: This article is an original article of "Lingyun Moment", please follow the reprinting requirements, and be sure to keep the author and source.