talks about how to choose load balancing equipment from the battle of the AlphaGo century.

fromAlphaGoThe Battle of the Century Talk about How to Choose Load Balancing Equipment

as for a game, "Alfa Dog" against Li Shishi has ended. As an artificial intelligence show, this craze has just hit.

we are admiring Google's artificial intelligence technology for "4:1"

this is the final result of Google's artificial intelligence "Afagou" (AlphaGo) machine in a game against Li Shishi, the former world's first go player and South Korea's nine-stage player.

, many people do not know that AlpaGo's victory is closely related to the maturity of distributed computing cluster systems. This is the result of different versions of AlphaGo games published by Google in Nature magazine:

the results of different versions of the AlphaGo game

, in the process of AlphaGo the distributed version and playing the stand-alone version AlphaGo, the winning rate of the distributed version reached 77%. From this point of view, the performance of the multi-machine system is much higher than that of the stand-alone system, and relying on the excellent distributed system, we can make the computing capability Scale-out expand without limit.

In fact, with the explosive growth of cloud computing and big data, distributed computing has attracted more and more attention. However, traditional services want to obtain the benefits of distribution without software transformation. Server load balancing technology is the first intelligent data exchange technology that has attracted the attention of industry users. It is centrally delivered to users through the integration of various new technologies. Load balancing equipment is also called application delivery equipment by some manufacturers, and more and more technology integration supports traditional business, cloud computing and big data business, forming application delivery center scheme and cloud application delivery platform scheme, and ensuring the smooth and reliable operation of data center.

However, load balancing, as a system-based product that integrates hardware devices and solutions, is not like aServeror PC can be distinguished by configuration parameters. In a lot of manufacturers load balancing equipment parameters, choose which product is suitable?

Cloud Load Balancing (Applicationdelivery) of the professional company Cormorant Technology Technical Director Luo Zong introduced, to introduce to you

several categories of load balancing products and how to choose professional load balancing hardware and the trend of corresponding load balancing technology in Europe and the United States in the future.

load balancing

First of all, we introduce the two types of basic classification of load-averaging equipment.

1,Link Load Balancing Device

2,server load balancing, also known as application delivery equipment, some manufacturers call it intelligent switches,

server load balancing can be divided

a,four-layer load balancing

B,seven-layer load balancing

According to the user needs to load the local server or load the multi-region data room is divided

a,Local Server Load Balancing (Server Load Balance)

B,global load balancing (Global Server Load Balance)

according to the product form is divided

a,Load Balancing Hardware Appliance

B,Load Balancing Container Server

c,Cloud Load Balancing (Virtual Edition)

so many products are called load balancing, how to distinguish and choose the products we need?

Link Load Balancing

must first determine whether the main application range of this product is link load or server load. The network application modes of these two scenarios are very different, which will directly affect the selection parameters.

in the link load, the selection of load balancing is more close to the ordinary network equipment. In general, the network shows high traffic, but the new is not high.

Therefore, in a link load scenario, throughput is the most important metric. In actual selection, users can be calculated according to the network egress bandwidth, actual peak traffic, and the maximum number of concurrent users (sessions) online.

In link deployment, load balancing does forwarding based on IP address and protocol type.four layersmode, so users can choose the right product model according to the throughput of four layers. If the equipment uses both the link load balancing function and the application load balancing function, the corresponding conversion is more difficult. In fact, the corresponding link load balancing equipment only involves the number of four layers of networks and is more inclined to regular firewall protection. The hardware built with embedded (non-X86) will be more stable and professional, and professional hardware link load balancing is used, you can avoid buying high-performance equipment once.Golden Wavefees, effectively protecting their investments.

, it should be noted that vendors have different ways of nominal throughput. Some label HTTP effective throughput, while others label UDP throughput. The nominal UDP throughput value will be higher, but it does not truly reflect the actual processing capacity of the product. Therefore, if you want to select products through UDP throughput, it is recommended to double or even triple the required value to calculate the final required device throughput value.

in the link load, the corresponding is the strength of traditional switch manufacturers, each manufacturer will generally use the backplane bandwidth as the performance standard of link load balancing.

because it uses embedded network hardware (low cost, low power consumption, high stability) and the same performance parameters, it only needs to pay 1/10-1/50 of the cost of traditional server load balancing (the cost performance of enterprise-level products is extremely high, and the cost of industry-level products corresponding to redundant power supply and other industries is increased, but it still has superior cost performance). In terms of the ability to process layer -4 data, the cost performance and stability far exceed that of X86 architecture, and the corresponding application load balancing will become more and more software in Europe and the United States (virtual machines will be incorporated into cloud platforms, which we will explain in later chapters), while link load balancing will become more and more hardware-based (border devices, hardware form is the best). There are several reasons why link load and server load need to be purchased separately.

1,costs: There is a huge difference between link load and server load cost. If the server load is selected according to link load parameters, funds will be greatly wasted.

2,hardware and software direction differentiation: Link load often emphasizes boundary capability. Professional users will tend to choose more stable and professional hardware, while customers with strong cost awareness will choose the functional modules of existing routers and firewalls to meet the requirements of link load,server load balancing will be concentrated in cloud data centers in the future. Virtualization is the inevitable trend of server load balancing, which is also the industry trend after the cloudification of data centers abroad.

3,their respective stability requirements: the stability of boundary equipment is more important, the server load focuses more on layer 7 data, and the link load only processes layer 4 data. if the server load is too large, the network is paralyzed due to the weak stability of the boundary equipment, and even the service interruption often outweighs the loss.

4,specialized division of labor requirements: the locations and requirements for customers to choose server load balancing and link load balancing are often very different. with the improvement of business requirements, users will not only distinguish server load balancing from link load balancing, but also separate the load of web server, middleware application layer server and database server, and use separate load balancing respectively, at this time, most of the load balancing considerations are the optimization of the application itself, and the link load function is no longer needed.

Application Server Load Balancing

, if link load balancing mainly understands throughput and concurrency, as well as the corresponding hardware architecture and cost performance, server load needs to be considered more. It first needs the number of users, data throughput capacity and comprehensive functions.

for server applications, its typical characteristics are high new, high concurrency, low traffic. An application can create a peak of 6000 new connections in just a few minutes, with more than 120000 concurrent connections, but the peak traffic is less300Mbps.

Therefore, in the choice of server load, we must first look at the number of new connections of the load product. This metric is like the mouth of a bottle: the larger the mouth, the higher the number of users the server can handle.

the bandwidth throughput capacity at the same time, because the consideration of comprehensive capacity is also a very important parameter in performance.

, however, many engineers will say, "I don't know how many new connections per second are in the application scenario. I can't get this data at all." This is a fact. Most of the time, the number of new connections to the business can only be counted through professional equipment.multi-userdoes not know the number of new connections for the current service.

what should I do? in fact, it can be calculated.

required equipment performance = number of concurrent users * number of business sessions * total number of business * 1.5 + number of future users * number of business sessions * total number of business

For example, although the number of new connections is not known, the number of users can be understood. For example, there are 2000 users in a certain business. According to the nature of the business, it is estimated that there will be about 500 people online at the same time. Then, it is estimated that the number of connections initiated by each user accessing the business is 20 (which can be viewed by grabbing packets) and a total of 10 similar business systems (generally, one person will not log in at the same time, so the corresponding parameter is set to 3). Finally, the new connection selection of load equipment of the system:

(500*20)* 3*1.5+1500*20*3=45K-135K

that is to say, first select a load device with a new number of not less than 45 K/S (corresponding to the required performance multiplied by 2 if it is a seven-layer performance). On this basis, consider the throughput bandwidth.

such as the following series of products

LCASeries Enterprise Edition

, it should be noted that the server load should consider the services opened on the load device. For example, for pure four-layer services, only four-layer new services need to be considered. If seven-layer services are opened, seven-layer new services need to be considered.

to add.

the difference between four-layer load balancing and seven-layer load balancing:

Layer 4 load balancing: through the target address and port in the message and the server selection method set by the load balancing device, it is determined that the finally selected internal server establishes a TCP connection with the requesting client, and then sends the data requested by the Client.

can be seen from the above figure: in the four-layer load device, the target address of the message sent by the client (originally the IP address of the load balancing device) is selected according to the rules set by the balancing device for selecting the web server, so that the client can directly establish TCP connection with this server and send data.

Layer 7 load balancing is also called content switching, that is, the final server is determined mainly through the truly meaningful application layer content in the message and the server selection method set by the load balancing device.

can be seen from the above figure that the seven-layer load balancing actually acts as a proxy server. We know that three handshakes are required to establish a TCP connection. However, to access the webserver, the client must first carry out three handshakes with the seven-layer load equipment, establish a TCP connection, and send the message information to be accessed to the seven-layer load balancing. Then the seven-layer load balancing selects a specific webserver according to the set balancing rules, then establish a TCP connection with this webserver through three-way handshake, and then webserver send the required data to the seven-layer load balancing device, and the load balancing device sends the data to the client; Therefore, the seven-layer load balancing device acts as a proxy server.

seven layers are responsible for balancing the advantages of the device:

make the whole network more "intelligent", it can forward requests for pictures to the picture server, requests for text to the text server, visits from sales users are automatically forwarded to the delivery sales server, and visits from financial users are automatically forwarded to the financial server. Therefore, many manufacturers of seven-layer load balancing are also called intelligent switches, and on the basis of intelligence, they will have many deep application delivery capabilities, suchunilateral TCP acceleration, buffer acceleration, SSL acceleration, SDN extension support, SSO unified authentication support, and so on.If the corresponding HTTP cache and compression are enabled, performance loss will also be caused. There is also SSL offloading. If a corresponding service is started, the parameters of SSL creation and SSL throughput (1K and 2k key encryption strength are also different) should also be considered., at this time, the final bandwidth performance determines the difference between high and low end of the device.

choosing a good application load product in terms of function will have the following important functions to pay attention

(1)seven-layer load algorithm needs to identify the source IP address, HTTP header, HTTP header source IP address, server Cookie, server Cookie source IP address, active Cookie, active Cookie source IP address, cookie hash, cookie source IP address hash, URL hash, HTTP Post header, HTTP query string hash, custom HTTP header and other information, and the algorithm has a richer calculation degree

(2)seven-layer load will have natural support and recommended configuration for professional manufacturers such as Microsoft Exchange, Microsoft Lync, Microsoft ADFS, Microsoft RDS Templates, VMware Horizon View, VMware Horizon Workspace, etc., eliminating the trouble of using load balancing optimization for traditional enterprise users.

(3)seven-layer load is also critical, because load balancing has become the core component of the network and the protector of server business continuity, integrity, and reliability. The security of the seven-layer load balancing needs to be reflected in two aspects: first, the security of the system itself, the software system should adopt independent research and development, without system loopholes and backdoors; Second, it can provide necessary network security protection to reduce the threat of direct attacks on the server, which can reduce the single point of failure caused by the serial stacking of devices in the server area. Professional load can support IPS,IPS defense library can be updated online; Effectively prevent DDOS and SYN Flood attacks, and support WAF (Network Application Firewall) to make websites more secure.

(4)help for business optimization,unilateral TCP acceleration, buffer acceleration, SSL acceleration, SDN extension support, SSO unified authentication support and other functions to optimize and accelerate the business is a very important consideration for the commercial version of load balancing capability, if these capabilities are not mature and perfect, users often prefer to choose free open source load balancing to solve temporary needs.

, like the above Internet financial platform, they also use open source pache to basically meet the requirements in terms of business capability before there are security loopholes and compliance problems. However, as users access more and more, information security problems become more and more prominent. Considering learning from the compliance of traditional financial enterprises, they use SSL acceleration/uninstallation capability to strengthen encryption strength at the boundary to improve the security level of access, at the same time, the load of the border web server is reduced, and the ability of unilateral TCP acceleration and buffer acceleration is used to improve the customer's access experience. The capabilities of IPS and WAF are also of great help to protect the security of their website applications. At the same time, when considering the data disaster recovery scheme in the future, the capabilities of global load and SDN and VPN can help them to take into account the comprehensiveness of application delivery more intelligently.

commercial load balancing for more than ten years of product precipitation and service capabilities and sensitivity to new technologies can help traditional enterprises to use the best and most professional IT products and achieve twice the result with half the effort.

the maturity of professional commercial load balancing virtualization capabilities, replacing the original load balancing of open source virtualization is also a simple thing to achieve overnight.

(5)network is mainly achieved through the redundant design of key components of the equipment, as well as the deployment of dual computers and clusters.

(6)'s ability to optimize the X86 platform requires higher performance products due to the higher CPU load requirements of the seven layers. Although the hardware platform has been developed to multi-core at present, there is a big gap in the application development of multi-core platform among traditional load balancing manufacturers in China, which seriously restricts the coordinated distribution and use of computing power among multi-core of load balancing equipment. Even F5, the leader of load balancing, has just transferred from embedded platform to X86 platform (just now we mentioned the benefits of link load embedded platform, but why is the X86 platform migrated, in fact, it is mainly for cloud computing services). Whether it can be optimized on X86 platform is especially important for seven-layer load balancing.

(7)'s support for cloud computing, with the continuous improvement of the number of concurrent services overseas, the high availability of users' key services is becoming more and more important. How can enterprises use server clusters more effectively and apply load balancing (foreign link load work because there are no domestic north-south interconnection, multiple internal and external networks and other issues, relatively speaking, they are all supported by boundary devices with professional link load balancing capabilities) has become an essential component of the enterprise's core business system. Virtualized load balancing becomes standard for every application. Besides the traditional hardware equipment, whether to provide various virtualization version support and optimization (especially the support of general-purpose vmware linkage version, etc.), public cloud version and container mode (traditional server becomes professional load balancing server) support must also be considered by far-sighted users.

(8)global load is also critical to application load balancing

if local server load balancing is to deliver multiple servers into "one server" through one load balancing, then the overall situation is to help users actually deliver multiple computer rooms into "one computer room" through multiple technologies. global load balancing provides high availability and disaster tolerance of business systems in multiple data centers, and is suitable for migration and switching of business systems in private clouds and public clouds, the quality of the global load determines the scalability of the local server cluster (how to intelligently and safely expand to private cloud, remote (disaster recovery) computer room, public (disaster recovery) cloud). Some unprofessional manufacturers claim to have the ability of global load only with simple intelligent DNS function. In fact, professional global load still has the ability to pay attention:

a,which algorithms are embedded in global load balancing, including polling, weight polling, fixed weight, load of business system, longitude and latitude of user access location, and proximity of user access location to data center.

B,whether the access policy can be set according to whether the data center site is a public site or a private site.

c,supports a custom recovery model when the site is unavailable: manual recovery or automatic recovery.

d,whether global load balancing supports session persistence

e,support seamless integration with DNS for automatic addressing

f,whether the built-in IP(IPv6 and IPv4) location database, and regularly update,

g,supports external deployment: Deploy devices outside the data center to provide high availability and disaster recovery.

h,supports custom regions and the mapping of custom IP addresses and regions. Realize the unification of multiple data centers in different places.

I,supports cluster deployment. Devices are deployed in each data center to provide global load balancing and serve as a logical device.

j,whether the cluster deployment mode supports automatic configuration synchronization between devices and unified kernel upgrade management of devices.

these capabilities determine whether multiple platforms and computer rooms can be delivered more intelligently into "one computer room" or even "one server" in the end ".

(9)localized users need localized service capabilities

, because the division of labor of foreign products is very detailed, some products are not enough in localization capability, such as localization big data audit capability, local address database update capability, single sign-on capability, application virtualization integration capability, multi-device centralized management and upgrade capability. These personalized localization capabilities are also important factors for users to choose professional manufacturers.

(10)manufacturers' professionalism in the use of load balancing and their understanding of the overall solution (cloud proximity, cloud application delivery center, disaster preparedness, SDN extension support, etc.) are also key to user selection. Why do traditional financial and industry users purchase commercial load balancing hardware and software in large quantities because of compliance and other needs, instead of choosing open source solutions, is considered from this professional perspective and professional services.

To sum up, if you really understand the meaning of the performance parameters of load balancing products, it is not difficult to select load balancing products. However, after considering the compliance and high requirements of enterprises, it is especially important to choose a suitable safe and powerful commercial product and a manufacturer with foresight and high professionalism.

LuccitechCormorant Technology Introduction

Company has 10 million registered capital, more than ten years of professional delivery and security product experience, the first domestic cloud application delivery product solution provider, and has a group of experts who are the first to participate in the overall solution of cloud computing at home and abroad. The company's headquarters and R & D base are located in Shenzhen, with marketing centers and service centers in Guangzhou and Shenzhen respectively, and offices in Nanning and Nanchang. The company's core staff from Microsoft, IBM, Lenovo, China Telecom, Telecom Research Institute and other first-line industry IT service companies. Cormorant Technology is committed to providing forward-looking and cloud-evolving expert-level industry application delivery services for industry enterprise customers through the cloud application delivery platform as an overall solution. So far, it has provided overall delivery solutions and services for many industry informatization leaders in China, and is a professional company aiming at helping Chinese enterprises provide global IT support solutions.

company is taken from China, which is often used to symbolize industrious and beautiful birds,cormorant, adhering to the spirit of IT services.

to become an IT framework to help traditional enterprises realize the Internet and help domestic enterprises and institutions realize leapfrogging,

is a professional information product leader that provides global IT support solutions for Chinese enterprises.

cloud application delivery experts, adhering to the spirit of IT services.

to help traditional enterprises to achieve the Internet leapfrog IT architecture

provides global IT support for Chinese enterprises