What is a cluster? Why use a cluster architecture?

What is cluster ？ Why use a cluster architecture ？

2022-06-09 16:58·

Finovy Cloud

One 、 What is cluster ？

In short , A cluster is a group of （ Several ） Independent computers , A large computer service system composed of high-speed communication network , Each cluster node （ Each computer in the cluster ） Are independent servers running their own services . These servers can communicate with each other , Collaboration provides applications to users , System resources and data , And managed in a single system model . When a user requests a clustered system , The cluster gives users the impression that it is a single independent server , In fact, the user requests a group of clustered servers .

for instance ：

Open Google , Baidu's page , It looks so simple , Maybe you think you can make a similar web page in a few minutes , But in fact , Behind this page is the result of thousands of server clusters working together .

To describe a cluster in one sentence , That is, a group of servers cooperate to do the same thing , These machines may need unified and coordinated management , It can be distributed in one computer room , It can also be distributed in multiple computer rooms in various regions of the country and the world .

Two 、 Why use clustering ？

（1） High performance

Important computing intensive applications in some countries （ Like the weather forecast , Nuclear test simulation, etc ）, The computer should have a strong ability to calculate and process . With the technology available all over the world , Even large machines , Its computing power is also limited , It is difficult to accomplish this task alone . Because the calculation time may be quite long , Maybe a few days , Even a few years or more . therefore , For this kind of complex computing business , The computer cluster technology is used , Dozens or hundreds of them , Even thousands of computers do calculations .

If you have one LNMP Environmental Science , Only service is needed at a time 10 Four concurrent requests , Then a single server must be faster than multiple server clusters . Only when the number of concurrent or total requests exceeds the capacity of a single server , Server cluster will show its advantages .

（2） Price effectiveness

Usually a system cluster architecture , Only a few or dozens of server hosts are needed . It is much cheaper than the special-purpose supercomputer which is worth millions of yuan at every turn . Under the same performance requirements , The computer cluster architecture is more cost-effective than the large-scale computers with the same computing power .

Early Taobao , The database and other core systems of Alipay use minicomputer servers with millions of yuan . Later, due to the high cost of use and maintenance and the geometric doubling of the cost of equipment expansion , Even become the bottleneck of expansion , Personnel maintenance is also very difficult , End use PC Server cluster replacement , such as , Combine the database system with a small computer Oracle Database migration to MySQL Open source database integration PC On the server . Not only did the cost drop , It is also easier to expand and maintain .

（3） Scalability

When the service load , When the pressure increases , The cluster system can be simply extended to meet the requirements , And will not reduce the service quality .

Usually , If the hardware device wants to expand the performance , Have to add new CPU And memory devices , If you can't add it , You have to buy higher performance servers , Take our current servers for example , The number of devices that can be added is always limited . If cluster technology is adopted , You only need to add a new single server to the existing cluster architecture , From the point of view of the customer visited , The system service is almost unchanged in terms of continuity and performance , The system has been upgraded unconsciously , Increased access , Easily implemented extensions . The number of nodes in the cluster system can grow to thousands or even tens of thousands , It's much more scalable than a single supercomputer .

（4） High availability

A single computer system will always face the problem of equipment damage , Such as CPU, Memory , a main board , Power Supply , Hard disk, etc. , If only one part breaks , The computer system may be down , Unable to provide services properly . In a clustered system , Although some hardware and software will still fail , But the service of the whole system can be 7*24 Hours available .

The cluster architecture technology can make the system continue to work in case of several hardware equipment failures , This minimizes system downtime . The cluster system improves the system reliability at the same time , It also greatly reduces the business loss caused by system failure , At present, almost 100% All Internet sites require 7*24 Hour service .

（5） transparency

A loosely coupled cluster system composed of several independent computers constitutes a virtual server . When a user or client program accesses a clustered system , It's like accessing a high-performance computer , Like a highly available server , Some servers in the cluster go online , Offline will not interrupt the service of the whole system , It's also transparent to users .

（6） manageability

The whole system can be physically large , But it's easy to manage , It's like managing a single image system . In an ideal situation , The software and hardware modules can be plug and play .

（7） Programmability

On a cluster system , Easy to develop and modify various applications .

3、 ... and 、 Common classifications of clusters

1、 Common classifications of clusters

Computer cluster architecture can be divided into the following categories according to function and structure ：

• Load balancing cluster , abbreviation LBC perhaps LB

• High availability clusters , abbreviation HAC

• High performance computing cluster , abbreviation HPC

• Grid computing cluster

Tips ：

Load balancing cluster and high availability cluster are common cluster architecture modes in the Internet industry , It is also the focus of our study .

2、 Common classifications of clusters

（1） Load balancing cluster

- Load balancing cluster provides more practical for enterprises , More cost effective system architecture solutions . The load balancing cluster can distribute the load pressure of access requests from many customers equally in the computer cluster . Customer access request load usually includes application processing load and network traffic load . Such a system is well suited to the pattern of using the same set of applications to serve a large number of users , Each node can bear the load pressure of access request , And it can realize the dynamic allocation of access requests among nodes , To achieve load balancing .

Load balancing cluster runtime , Generally, customer access requests are distributed to a group of servers on the back end through one or more front-end load balancers , So as to achieve high performance and high availability of the whole system . Generally, high availability cluster and load balancing cluster will use similar technology , Or it has the characteristics of high availability and load balancing at the same time .

The role of a load balancing cluster is ：

• Allocate user access requests and data traffic （ Load balancing ）

• Maintain business continuity , namely 7*24 Hourly service （ High availability ）.

• be applied to Web And database and other server services

Typical open source software for load balancing clusters includes LVS,Nginx,Haproxy etc. . As shown in the figure below ：

Tips ：

Different services have a switching time of several seconds ,DB The business is obviously better than Web Service switching time .

（2） High availability clusters

Generally, it refers to when any node in the cluster fails , All tasks on this node will be automatically transferred to other normal nodes . This process does not affect the operation of the entire cluster .

When a node system in the cluster fails , The cluster service of the operator will react quickly , Allocate the services of the system to other working systems in the cluster . Considering the fault tolerance of computer hardware and software , The main purpose of high availability cluster is to make the overall services of the cluster as available as possible . If the primary node in the high availability cluster fails , Then the backup node will replace it during this period . The standby node is usually the mirror of the primary node . When it replaces the primary node , It can take over the master node completely （ Include IP Address and other resources ） Provide services , therefore , Make the clustered system environment consistent for users , That is, the access of users will not be affected .

The high availability cluster makes the running speed and response speed of the server system as fast as possible . They often use redundant nodes and servers running on multiple machines to track each other . If a node fails , Its replacement will take over its duties in a few seconds or less . therefore , For users , Any machine in the cluster goes down , Business will not be affected （ In theory ）.

The role of a high availability cluster is ：

When a machine goes down , Another machine takes over the downtime of the machine IP Resources and service resources , Provide services .

It is often used in applications that are difficult to achieve load balancing , Such as load balancers , Master database , Between primary storage pairs .

Open source software commonly used in high availability clusters includes Keepalived,Heartbeat etc. , The frame composition is shown in the figure below ：

（3） High performance computing cluster

High performance computing cluster is also called parallel computing . Usually , High performance computing clusters involve parallel applications developed for clusters , To solve complex scientific problems （ The weather forecast , Oil exploration , Nuclear reaction simulation, etc ）. The high-performance computing cluster is like a supercomputer , The supercomputer is internally composed of tens of thousands of independent servers , And communicate on the common messaging layer to run applications in parallel . In the production environment, the task is actually cut into cakes , Then it is distributed to the cluster node for calculation , The result is returned after calculation , Then continue to calculate the new task , So back and forth .

（4） Grid computing cluster

Because it is seldom used , Skip here

hot tip ：

In Internet websites , Load balancing cluster and high availability cluster are commonly used

Four 、 Introduction and selection of commonly used cluster software and hardware

1、 Common cluster software and hardware products in enterprises

The open source cluster software commonly used by Internet enterprises includes ：Nginx,LVS,Haproxy,Keepalived,heartbeat.

The common business cluster hardware used by Internet enterprises are ：F5,Netscaler,Radware,A10 etc. , The working mode is equivalent to Haproxy Working mode of .

TaoBao , Catch up network , Sina and other companies have used Netscaler Load balancing products . Cluster hardware Netscaler The product diagram of is shown in the figure below ：

Cluster hardware F5 The product is shown in the figure below ：

2、 How to select cluster software and hardware products

When business matters , The technical force is weak , And hope to pay for products and better services , You can choose hardware load balancing products , Such as F5,Netscaler,Radware etc. , Most of these companies are traditional large non internet enterprises , Like a bank , negotiable securities , Finance and BMW , Benz, etc

For portals , Most of them use software and hardware products to share the risk of a single product , Such as Taobao , tencent , Sina, etc . The financed enterprises will buy hardware products , Such as ganji.com .

Small and medium-sized Internet enterprises , Because there is no profit to be made at the initial stage or the profit is very low , Will hope to solve the problem by using open source and free solutions , Therefore, special operation and maintenance personnel will be employed for maintenance . for example ：51CTO etc.

By comparison , The cost of commercial load balancing products is high , Good performance , A more stable , The disadvantage is that it cannot be redeveloped , Open source load balancing software has high requirements on the ability of operation and maintenance personnel , If the operation, maintenance and development ability is strong , Then open source load balancing software is a good choice ,

The current Internet industry tends to use open source load balancing software .

3、 How to select open source cluster software products

The websites of small and medium-sized Internet companies are accessed concurrently and the total number of visits is not very large , Preferred Nginx Load balancing , The reason is that Nginx Load balancing configuration is simple , Easy to use , Safe and stable , The community is active , More and more people are using it , Become a popular trend , Another similar product for load balancing is Haproxy（ Support L4 and L7 load , Equally good , But the community is not as good as Nginx active ）.

If you want to consider Nginx High availability of load balancing , Preferred Keepalived Software , The reason is that installation and configuration are simple , Easy to use , Safe and stable , And Keepalived There are other highly available software that serve similar services Heartbeat（ It's complicated to use , It is not recommended for beginners ）

If it is a large enterprise Internet company , Load balancing products can use LVS+Keepalived Do layer 4 forwarding at the front end （ It is generally active / standby or active / active , If you need to extend, you can use DNS Or front-end use OSPF）, The back-end using Nginx perhaps Haproxy do 7 Layer forwarding （ It can be expanded to hundreds ）, Then there is the application server , If it is load balancing and high availability of database and storage , Suggested choice LVS+Heartbeat,LVS Support TCP Forward and DR The model is very efficient ,Heartbeat Can cooperate with drbd, Not only can VIP Handoff , Block device level data synchronization can also be supported （drbd）, And the management of resource services .