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

The core requirement of the internal system of the data center is "stable and reliable", which means that the system has the ability to provide continuous and reliable services and run without failure for a long time; second, when a fault occurs, it has the ability to locate quickly and troubleshoot in time, and the fault range does not spread.

The separate deployment mode makes the system independent from the cloud platform system, avoiding computing and storage scrambling for physical resources such as CPU/ memory / network, once the resource demand of one side is exhausted by the other side, which affects performance and produces ripple effect in the whole infrastructure; and in the super-convergence deployment mode, when the cluster scale is large, the probability of network, hard disk, server failure will increase. As well as data re-deletion, compression, encryption and erasure code and other functions, fault self-repair and data function implementation will consume certain system resources, resulting in performance degradation and jitter and other problems.

Advantages of split deployment over superconvergence:

If this view is not from a certain manufacturer or supplier, it is too biased. I think the two technologies are suitable for different scales. Small and medium-sized (including platform scale, including human resources scale) has obvious advantages of super-integration, and large-scale distributed storage has more advantages. Good design, proper balance is the key, there is no one-sided absolute advantage.

A super-integrated deployment model is recommended.

1. In terms of cost, each super-integrated server can be used as both a computing resource and a storage resource, with the highest performance-to-price ratio.

2. In terms of performance, distributed storage systems generally occupy only about 10% of the server resources, and 90% of the performance can be used. And at present, the CPU performance of X86 server is getting higher and higher, if it is not used idle, it is a pity that the utilization is too low.

3. In terms of management, every X86 physical device is the same, which is convenient for horizontal expansion and management. each additional node increases computing resources and storage resources at the same time, which is beneficial to management. At present, Baidu and Ali's cloud data centers basically do this.

I. the concept of super-integration

At present, superfusion (Hyper- Converged) does not have a strict standard definition, and each manufacturer and institution has its own definition, which also shows that superfusion is still in rapid development and evolution, and has not formed a unified standard specification. "Super" in super convergence corresponds to "Hyper" in English, especially virtualization, and corresponds to virtualized computing architecture, such as KVM, XEN, Hyper-V and so on. This concept originated from Nutanix and other storage manufacturers using the computing storage convergence architecture adopted by Internet manufacturers such as Google/Facebook in the virtualized environment to provide enterprise customers with a computing storage integration product or solution based on X86 hardware platform. According to this concept, neither database all-in-one nor big data all-in-one can be super-integrated, because applications such as RAC/Hadoop do not run on virtual machines. In addition, the most fundamental change in the super-convergence architecture is storage, from centralized shared storage (SAN/NAS) to software-defined storage, especially distributed storage.

"Convergence" in superconvergence means that computing and storage are deployed on the same node, while providing computing and storage capacity. Generally, fusion can be divided into physical fusion and super fusion, which is a subset of fusion. In the physical fusion system, computing and storage can still be two independent components, without direct interdependence, and share the physical resources of the host. Super convergence is different from physical convergence in that it focuses on virtualized computing, and computing and storage are closely related. Storage is controlled by controller virtual machines (Controller VM,CVM) rather than physical machines, and decentralized storage resources are formed into a unified storage pool for creating users' application virtual machines. The comparison between physical fusion and superfusion is shown in. For performance reasons, super-convergence architectures usually need to Pass Through the host physical devices to the controller virtual machine CVM.

The super-fusion architecture is based on standard and general-purpose hardware.

Platform, through the software definition to achieve computing, storage, network convergence, to achieve virtualization as the center of the software-defined data center technology architecture. To judge whether a system adopts a super-fusion architecture, it is mainly based on the following points:

(1) complete software definition. Independent of hardware, using commercial standard hardware platform (such as X86), completely using software to achieve computing, storage, network and other functions.

(2) complete virtualization. Taking virtualization computing as the center, computing, storage and network are all managed and scheduled by the virtualization engine, and the software-defined storage is managed by the virtual machine controller CVM.

(3) completely distributed. Horizontal expansion of the distributed system, computing, storage, network on-demand dynamic expansion, the system does not have a single point of failure, using distributed storage.

II. Development and cases of super-integration architecture

In the next 5 to 10 years, a new generation of data center infrastructure is moving towards software definition and superconvergence, and SAN/NAS storage is gradually replaced by software-defined storage. Driven by software-defined storage (SDS), hyperconvergence will become the core of data center infrastructure and the future technology development trend of software-defined data center (SDDC). Data centers based on a variety of complex devices will eventually be normalized into a system architecture with general-purpose servers and interconnected networks. Deploy key software on these general-purpose servers to realize computing and storage resources by means of virtualization. Then deploy the application on the basis of these virtualized resources and complete the specific functions.

The deep convergence of storage, computing and network is the major trend of IT infrastructure development in the future. Therefore, super-convergence architecture has become the first choice for enterprise customers, accelerating the transformation of business systems from traditional architecture to cloud computing architecture. According to the statistical report of IDC, the regulatory model of the global super-integration market is expected to grow by 94% in 2016. Gartner, another market analysis firm, expects the global superconvergence market to exceed $100 billion in 2019, with about 30 per cent of storage arrays installed in enterprise data centers deployed in software-defined storage or super-converged integrated system architectures based on x86 hardware series.

Since 2016, China's super-integration market has continued to heat up, and more and more super-integration teams have gradually developed. According to a report released by Gartner, a market research and consulting company, on the trend of China's super-convergence market in November 2016, established server and storage companies such as Cisco, HP, Dell, EMC and NetApp have shifted their technology and product strategies to super-convergence. According to the report, China has become the fastest growing market for super-converged infrastructure in the world.

Domestic H3C, Huawei, Nutanix, Lenovo, SmartX, Shenxin and other manufacturers quickly followed the international mainstream super-integration solutions.

At present, the main application cases of super-fusion architecture in China are as follows:

Relevant government agencies: China Inspection Publishing House, China Securities Regulatory Commission, China Internet Information Center, China Dalian Municipal Government, Qinghai Water Conservancy, Guangzhou Local Taxation Bureau Database Virtualization, National Health Planning Commission Cloud data Center, Hubei Provincial Public Security Bureau Audit platform, Xiamen Public Security Bureau Police Cloud platform, Supreme Law Judicial Statistics Management platform, Shenzhen Customs Business system, Guangdong Maritime Safety Bureau Intelligent Maritime platform China (Xi'an) Silk Road Research Institute, the first affiliated Hospital of Wenzhou Medical University, etc.

Financial industry: China Merchants Bank, Bank of China Securities, Nanjing Securities, CITIC Bank, etc. Manufacturing industry: Dongfeng Honda, Baoshan Iron and Steel, China Railway Resources Group Co., Ltd. IT enterprises: Kelu Electronics, Lenovo Group, China Unicom Voyun, China Telecom, etc. Education industry: China University of Geosciences, China University of Science and Technology, Nankai University, Beijing University of posts and Telecommunications, Beijing Institute of Foreign Affairs, Shaanxi Provincial Institute of Administration, etc.

Third, the advantages of super-integration architecture

The reason for the rapid development of super-fusion architecture is that it has significant advantages and can bring high customer value. The super-convergence architecture realizes the unified management and scheduling of computing, storage, network and other resources, has a more flexible scale-out ability, and can bring optimal efficiency, flexibility, scale, cost and data protection to the data center. The use of computing storage super-integrated platform replaces the traditional server plus centralized storage architecture, which makes the whole architecture clearer and simpler, and greatly simplifies the design of complex IT systems.

One of the key features of the super-fusion architecture is easy to scale, minimum deployment, and on-demand expansion. Superfusion manufacturers claim that the maximum cluster size also varies widely, ranging from tens to thousands of nodes, usually starting from 3 nodes. In super-convergence, computing capacity, storage performance and capacity are expanded synchronously, which can not meet the expansion of individual capabilities in reality. Some manufacturers also have requirements for the expansion of minimum units, and the flexibility of expansion will be limited.

After the cluster reaches a certain scale, the complexity of the system architecture will increase nonlinearly, the cluster management will become more difficult, and the probability of hardware failure and self-repair will be greatly increased. Therefore, the super-convergence architecture generally does not recommend the construction of large clusters, if the business allows to build as many smaller clusters of appropriate size as possible, or to build failure domains or sub-resource pools in large clusters.

The physical device is transmitted to the control virtual machine, which increases the complexity of deployment configuration management. Computing and storage have different requirements for hardware platform, and the integration will increase the complexity of compatibility verification to a certain extent. Under the super-converged architecture, management, computing, storage and high availability usually require the configuration of independent virtual networks, and the network configuration will be more complex. At the same time, the allocation, isolation and scheduling of shared physical resources is also an additional complexity. If a fault occurs, it becomes more difficult to track, debug, analyze and diagnose the problem.

The storage logic unit under the super-fusion architecture already has many functions that advanced storage used to have, but in data protection, replication, disaster recovery, and high availability, these requirements related to data storage are that super-fusion manufacturers will not pay attention to and cannot pay attention to. At the same time, the limitations of the architecture itself bring about a relatively narrow range of user choices, so you can't fill super-large computing centers with super-fusion architecture. The massive growth of users for structured and unstructured data, the heterogeneous demand for data storage is not suitable to be solved by hyperconvergence. Many users had hoped to expand the computing and storage distribution, or the proportion of computing to storage is out of balance. For example, the demand for computing resources is very large, and the demand for storage resources is very small, or vice versa. It is quite bad to leave everything to the super-integrated architecture design.

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.