Shulou(Shulou.com)06/01 Report--

This article is based on the speech delivered by Li Mingjun, General Manager of Huasheng Tiancheng Cloud Computing R & D and Product Center, at the "TF Chinese Community Establishment and the first full meeting". For more information about the meeting, please reply to "Inaugural meeting" on the official account.

Li Mingjun, General Manager of Huasheng Tianchengyun Computing R & D and Product Center

I am very glad to have the opportunity to share with you Huasheng Tiancheng's experience in the landing of cloud computing open source networks.

We came into contact with Tungsten Fabric in the first half of 2019, and it has been more than half a year now, and we are very pleased to see such an excellent solution being put into the community.

Enterprise user needs: open, heterogeneous, scene-oriented

In the past decade, we have seen cloud computing change from a concept to a mainstream architecture. In the process, our customers' requirements for cloud computing technology architecture, as well as functional expectations, are also changing.

For the medium and large enterprise market, the demand has evolved from the initial heterogeneous to the later heterogeneous hybrid, to today it has become the management requirement of heterogeneous hybrid multi-cloud-at the infrastructure level, there are desktop clouds, various types of resource pools that exist in the form of virtualization, as well as various public cloud resource pools, and public cloud applications have all entered the IT environment of medium and large enterprises. Enterprises need to find an integrated infrastructure that directly serves the business in such a heterogeneous, hybrid, cloudy environment.

This brings a very real need, and we sum up three words: open, heterogeneous, and situational.

How to understand openness? Corresponding to openness, in the early days, many proprietary solutions, or solutions led by a single vendor, brought about restrictions on functional expansion and business cooperation.

Heterogeneity occurs at many levels, such as historical IT architecture and existing applications and IT infrastructure, or today's many types of cloud computing infrastructure services, including private and public clouds.

In addition, many of the systems built by customers ten years ago may still be running and carry some very important businesses. how to integrate these system services with the emerging Internet and Cloud-native architecture and infrastructure is an urgent problem to consider.

For medium and large enterprises, how to integrate all the IT capabilities of users in the data center to form a cloud service has become a very urgent need for enterprises. We have encountered a lot of such customers, whether in the power industry, operators, or in the financial industry.

The third is the scene, no matter how high-end the technology is, the first question for customers to ask is, what is the promotion of the new technology to the business scenario? Which business scenarios can directly improve business efficiency? The first thing that enterprises need to think about when they invest in IT is how to bring help to the business.

When we implement some cloud solutions, we will emphasize which ones can be associated with the customer's specific business scenarios. A key idea here is how I can relate the very rich and diverse things in the cloud to the business and my applications.

These are the three aspects of demand that we see.

In this context, we can see that the network, as the skeleton and nerve of cloud computing, is very important. Only through the network can we provide connection and support to the heterogeneous environment and manage these heterogeneous, hybrid and cloudy resources.

The "three axes" of enterprise cloud management platform

We respond to the above needs and challenges mainly from three aspects.

The first is heterogeneous multi-cloud infrastructure resource management. In terms of hybrid cloud management, we see several mainstream technical solutions for private clouds. For example, OpenStack, VMware, which has a lot of resource pools in the commercial market, POWER as a minicomputer resource pool, and xCAT, a bare metal management tool, is the mainstream resource pool K8s in cloud native applications. In the case of public cloud, it is mainly Aliyun, Huawei Cloud and Tencent Cloud in China, and then AWS also has more and more customers in China.

The heterogeneous computing storage network, including the heterogeneous management of technical resources and technical software, is actually aimed at the heterogeneous hybrid and cloudy infrastructure. We need to provide a unified entrance and a unified resource pool management tool to form a service directory of all resources.

Then on top of this, let's see how to face the scene of the customer, and how to provide direct help to his business, in which we have two major application functions: one is the scene of Tencent App Center.

Compared with traditional cloud service products, we only provide some virtual machines and basic network services in the cloud. In basic storage services, you can apply for a virtual machine, and then create a virtualization-based operating system, in which you can apply for a service IP and then hang it on the network of a storage pool. We think this is a mainstream cloud management product in the past few years. We define it as a cloud product that can provide services at the resource level.

Now we have the conditions to raise it to a level that supports the enterprise's business. In this scenario, Tencent App Center, we can define a set of services in cloud computing services, which are interrelated, regardless of computing resources, storage resources, network resources, and business systems based on the operation of these technical resources, such as a financial mutual fund platform, there may be dozens of virtual machines distributed in various network security domains. There are a variety of operating system requirements in these virtual machines, including Linux, UNIX or Windows, and storage requirements are also based on front-end applications and terminal applications.

How to orchestrate all the required IT resources in a business scenario and be able to deploy them automatically is a very big challenge. We have done a lot of landing practice in this respect.

There is also a business-oriented operation and maintenance center. Several previous experts also said that three-part resource management and seven-part operation and maintenance, in fact, in the entire multi-cloud architecture, the application-oriented business scenario is faced with a big problem, that is, how to better operate and maintain these resources under a business scenario.

In traditional architecture, when a business is online, or when a business is associated with a specific resource, there are two very important diagrams: one is the application architecture diagram, which describes the business system to be deployed from the perspective of the application architect; the other is the physical deployment diagram, which is used by the data center team to understand the physical layout of the business system.

In the traditional IT management mode, the application architecture diagram connects the development department of the application, and for financial institutions, it is the development and test center. The physical deployment diagram is made by the data center team, and a variety of basic computing, storage, network, and physical deployment roadmaps are made when the application middleware is landed.

In the absence of a business-oriented cloud management platform, we usually experience many cross-team communications, from development and testing to server administrators, storage administrators, data center administrators, and the entire operations team. in order to slowly evolve from the application architecture diagram to a physical deployment diagram, we can land in the data center and put a business online for operation and maintenance.

In our new generation of cloud management platform, everything has become an online automation mode.

First of all, we can create an application in the cloud management platform, and then add logical units familiar to the application architect. By adding services to these logical units, we can associate the dependencies between some components. These actions are formed by a graphical drag and drop online, which completes the logical definition of the business system.

Next we will define the logic, which is equivalent to the process of instantiating the logical architecture diagram by the data center architect, and we will configure each level by associating and configuring the resources of each logical unit with the physical resource pool at the back end.

For example, in this step, we are configured with a front-end deployment unit and run into a KVM virtual machine, which virtual machine runs in which resource pool in which network security domain, and finally forms the definition of the virtual node. The definition of the operating system in the virtual node, that is, its user group, its disk volume group, and its network security policy definition, will be completed on this node.

After each node is completed, the physical deployment diagram corresponding to the traditional manual era is finally formed. The whole process supports the hybrid architecture of private cloud and public cloud, as well as the heterogeneous hybrid multi-cloud environment of multiple cloud computing resource pool vendors. In such an environment, a business system is landed.

In this process, we can see that whether it is the KVM resource pool, the OpenStack resource pool supported by OpenStack, or the commercial VMware resource pool, when it corresponds to the interior of a single resource pool, its management tools are relatively sound. For example, VMware provides a very complete set of tools for virtualization management, and OpenStack has slowly developed a lot of open source tools. Within a single resource pool, these things are relatively easy to solve.

Unified Management and SDN selection

Once multiple types of resource pools are involved, that is, the isomerized multiple technology stacks we just defined, when there are private clouds, public clouds, and resource pooling technologies including multiple clouds, the biggest headache comes. In fact, it is how the network can string these diversified services together so that they can be related to each other and call each other.

Before we came into contact with Tungsten Fabric, we actually used a lot of brains and done a lot of development on the side of the cloud management platform, which means that we will implement some layer-2 and layer-3 solutions on the side of the cloud management platform to solve the problem of interconnection and interworking of internal networks in multiple resource pools.

Here we have two layers of work. First, as the entrance to the entire data center, the cloud management platform needs to solve not only the interconnection between the IT resource pools of multiple manufacturers, but also the interconnection of its minicomputer and its physical servers within a single SDN solution. What is the interconnection solution on the side of its virtual switch? It is necessary to adapt according to each vendor's SDN solution, and realize the management of all heterogeneous resources under a certain type of SDN solution in the cloud management platform.

Then to the second step, there are some large customers will have a variety of SDN solutions, we see the most there are three manufacturers of SDN solutions. At present, every solution in the market is incompatible with each other and must be self-contained. It is almost impossible for you to have one OpenStack and two SDN controllers. In this case, we need to do an overall cloud computing solution, consider the interconnection of multiple SDN solutions on the cloud management platform side, and carry out the functional development and design of the second and third layers.

Today we see that such a framework as Tungsten Fabric has actually solved these two problems that we must pay attention to in the landing of large-scale cloud computing to a great extent.

For the interconnection between heterogeneous resource pools of a single SDN, the framework of Tungsten Fabric actually provides almost the most complete combination in the industry, including the support of the more mainstream OpenStack + KVM system, and K8s + docker, and even the commercial VMware, which provides support for their own virtual switches and virtual networks at the resource pool level through a single SDN controller.

Another thing we are optimistic about is that it is relatively advanced in the market in terms of interconnection between several public cloud networks, including support for mainstream foreign public cloud vendors such as AWS. I believe that the interconnection of domestic public cloud networks such as Ali, Tencent and Huawei can also be realized quickly through the efforts of the community.

From this point of view, no matter for a variety of technology stacks in private clouds and multiple public cloud technology stacks, the TF framework has achieved the most complete support, satisfying the interconnection of all networks in heterogeneous hybrid multi-cloud environments under the condition of using a single SDN. I think TF is doing the best now.

Why we need to multiple SDN framework, spend a lot of energy to do integration, because the past a single SDN solution they are not open, an IT manufacturer's program only supports their own switch, virtual switch management, through the TF open source framework, these problems are partially solved, through the development of the community, there may be no need for more than SDN controller "group war" in the future.

In fact, no matter the management of bare metal, the management of virtual resource pool can be managed through an open platform recognized by everyone, so the network framework in our new generation of cloud management platform will do the interconnection of various device layers based on TF. We think this is a better solution.

This is a comparison of several SDN solutions that are common in the market today based on our experience.

First of all, in the Chinese market, we see several major SDN solution providers, including Huawei, Huasan, Cisco and Juniper Networks, which are not listed for commercial products.

Then we made a comparison from three dimensions.

First of all, large and medium-sized enterprises are more concerned about the support for heterogeneous resource pools, TF has a very leading advantage in this respect, he can now support OpenStack very well, our first cooperation, in fact, is to integrate our own OpenStack products with TF, all aspects of functions, we are also joint to do some customer project demonstration, the final effect is still good.

Then there are more mature solutions for VMware, bare metal, K8s, the public cloud is also a relatively early market, support for AWS has been realized a long time ago.

In addition, TF is an open source version, with the advantages of the community, we can not only rely on the power of manufacturers, whether colleges and universities, or the strength of our manufacturers, we can put very TF not to do the function, based on the project scenario, based on our user scenario can be done quickly.

For example, we are more urgent, the expansion and management of some functions such as underlay, as well as the support of southbound protocols. For underlay, we can see that the main problem now is that the support for physical networks is relatively weak, most of which are based on their own hardware system. In this case, in fact, customers need to have multiple SDN, because sometimes network equipment is purchased in batches, or because of various business games, there are multiple SDN solutions in a customer data center, which becomes a fact.

So if we have a relatively open framework, we can easily support the Underlay management of different manufacturers, we believe that Tungsten Fabric will be a better alternative, in fact, customers do not need to have more than one SDN solution, if there is a program on this side of the cloud management platform, it is very easy to achieve the management of such functions of underlay, in fact, the necessity is greatly reduced.

The case of operator and Government Cloud

Next, I will tell you two cases.

This is what we landed in a provincial company of China Mobile to manage the cloud resource pool of the entire production environment. In such an architecture, the entry point for customers to use IT is our heterogeneous hybrid cloud management platform. At present, we are divided into three areas in this environment.

One is the SDN area, which is used to manage resource pools for OpenStack, VMware, bare metal, and so on. For non-SDN areas, due to network configuration problems, we have retained a non-SDN area, which mainly supports x86 bare metal and POWER servers. For VMware, we have a directly connected area of VMware to manage the resource pool for VMware.

Customers increasingly want to transfer their workload to the SDN area to achieve flexible management of the network. The value that SDN brings to customers is to decouple the resources of IT from the network.

A painful problem before is that when migrating across networks, VMware resources, or KVM resource pools, are always limited to some network configurations, including manual VXLAN configurations, so they cannot be automated well. With the function of SDN, especially after integration with the cloud management platform, it can realize the migration of virtual machines across network domains and become very automated.

The second project case is the government cloud of a municipality directly under the Central Government, which is a large resource pool based entirely on open source technology. In this project, there are 540 x86 physical servers, plus 20 POWER servers, which can provide 290 million TPMC processing capacity, all under the management of SDN. Whether it is computing or storage, it is a heterogeneous environment.

The value of SDN in this project lies not only in the migration of virtual machines across resource pools, but also in the decoupling of physical devices and networks. In addition, customers also implement virtual routers, virtual firewalls, load balancers and other functions based on SDN, all under the orchestration control of the SDN controller, and then form a unified service directory through the cloud management platform.

In fact, we have seen that starting from 2018, the landing of SDN has basically become a must in some large-scale projects in enterprises. In the past, due to the limitations of the single-manufacturer SDN scheme, it was a headache in terms of equipment co-tuning and function integration. Today we are very pleased to have such a TungstenFabric-based solution, which brings great convenience to the integration of large-scale enterprise solutions and the overall functional design of cloud computing solutions.

Thank you!

Follow Wechat: TF Chinese Community

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.