Disaster recovery series (III) -- cloud network disaster recovery construction

The network is part of the infrastructure , Network disaster recovery construction is an important indicator of data center acceptance . Imagine that there is a single point in the network link of a data center , Just like a city where all roads are one-way streets , In the event of a traffic accident , Small will lead to road congestion , Large will lead to traffic paralysis in the whole city .IDC Time , The business is less involved in network disaster recovery , It mainly depends on the degree of data center network disaster recovery construction ; When it comes to the cloud age , After the cloud service provider products the underlying network capabilities , Cloud customers are more involved in network disaster recovery , Improve business stability . This article gives an overview of cloud network , Cloud network disaster tolerance complexity and typical cases to introduce cloud network disaster tolerance construction .

1. Cloud Network Overview

Cloud network overview is mainly divided into cloud service provider infrastructure network architecture and cloud products , Let cloud customers have a deeper understanding of cloud network , Make good use of cloud network .

1.1 Cloud service provider network architecture

This section focuses on the following issues from the perspective of business disaster recovery construction ：

1） Are the underlying cloud networks in different availability zones of cloud service providers completely independent ？

Take Tencent cloud for example , When building a data center in a usable area , The bottom network and the power of the computer room shall be unified ; So different zones , The underlying networks are completely independent , Whether at the data level or the control level , Are completely isolated without intersection .

2） How long is the network delay in different availability zones in the same region ？

Take Tencent cloud for example , When selecting the computer room address in the same region , Distance greater than 60 km , It is required that the delay of different availability zones is less than 3ms, To meet the basic needs of cloud customers for local disaster recovery construction . When considering disaster recovery construction for cloud businesses , Virtual machines in different availability zones interact with each other PING To evaluate the specific delay . Take Guangzhou as an example , As a reference ：

Network delay in different availability zones in Guangzhou

3） Network stability between different availability zones

Take Tencent cloud for example , The WAN is connected through BR Core router interworking , Metropolitan area network DR Core router interworking architecture , As shown in the figure below ：

• Wan stability , The links between different regions provide redundancy for multiple optical fibers , At the same time, different regions form a ring , If part of the optical fiber of the direct link is interrupted , Priority is given to load recovery , In extreme cases , All direct links in different regions are interrupted , Restore service through the loop , Ensure business stability .
• Metropolitan area network stability , Different zones give priority to through trains to reduce network delay in different zones ; At the same time through “ Four fibers and three routes ” To improve the automatic self-healing capability of optical fiber interruption （ Less than 50ms）. If DR Part of the optical fiber of the through train is interrupted , adopt “ Four fibers and three routes ” To quickly restore business ; If DR All optical fibers are interrupted between , adopt BR Link to restore service .

wide & Metropolitan area network architecture diagram

4） How to deal with abnormal extreme conditions of the entire public network

Take Tencent cloud for example , Different regions will connect with the current ISP; For example, Guangzhou public network is abnormal , Cloud scheduling capability through intranet , Combined with network load capacity , From different regions ISP Visit the public network . for example 2019 year 3 month , The public network of Shanghai Telecom is abnormal , The cloud side has been used from fault discovery to recovery 2 minute . See http://www.etudu.com/?id=67.

1.2 Cloud network products

For cloud network products , From the business flow dimension, it is mainly divided into ：

Flow trend	Corresponding products	Disaster recovery construction
North South flow	Load balancing (CLB)、NAT gateway 、 Elastic public network IP（EIP）、anycast IP	1. How much life there is in the same city , Avoid cross zone traffic 2. Load balancing public network CLB Have the ability to cross AZ Disaster resilience 3.NAT The gateway is bound to multiple EIP, Increase the number of connections
East West flow	Private line access 、 Peer to peer Links 、 Cloud networking 、VPN、private link	1. For sensitive business, it is recommended not to vpn Get through 2. Hybrid cloud leased line access disaster recovery scheme , Meet each other 3.1 3.VPC It is recommended to adopt cloud networking for network interworking , Ensure simple network maintenance , The network architecture is clear .

2. Network disaster recovery complexity

Disaster recovery construction in the same city or in different places , There are three main factors at the network level ：

1） Cross region or cross region network delay , Impact on upper level business .

Network delay , The means to optimize the infrastructure are very limited , After all, it is limited by the actual physical distance and the speed of light . If the service is sensitive to network delay , It is usually to add middleware or buffer layer to reduce latency .

2） Disaster tolerance capability of cloud infrastructure across regions or regions .

Generally, the data centers of cloud service manufacturers have disaster tolerance capability , Here, it is suggested to choose large factories .

3）IDC To the cloud network high availability construction .

Hybrid cloud disaster recovery mode , Here, consider IDC And the disaster tolerance of the line on the cloud , It is generally recommended that two dedicated lines be connected to different POP Click to build disaster recovery ; At the same time establish VPN perhaps GRE Public network escape routes to restore business in an emergency .

3. Network disaster recovery cases

3.1 Public network CLB disaster

Public network CLB The multi availability zone capability has been launched , But you need the account disclosure support . If the stock public network CLB It is a single zone , It is recommended to upgrade to multi availability zone . Currently, smooth upgrade is not supported . Specific process ：

1. Need to purchase new multi availability zones CLB, Binding back end RS,

2. Cutting flow gray to multi availability area CLB, After normal business , Cut off all flow

3. Observe the flow of single availability zone , When there is no traffic or number of links , Official Downline .

Be careful ： For the entrance VIP Write about death , The cost of client upgrade is high .

CLB Multi zone purchase page

3.2 Hybrid cloud network disaster recovery

Hybrid cloud network disaster tolerance is divided into two parts ：

1）idc And the cloud machine room , The main lines are divided into special lines and VPN. Special line is the main line , Different POP Point access ;VPN Supplemented , The most emergency escape route , At the same time, pay attention to the cloud vpn The maximum bandwidth bearer of the gateway is 1G, If business requirements are not met , It is recommended to use GRE The scheme acts as an emergency channel .

2） Cloud side gateway disaster recovery , Mainly for dedicated line access , Through the cloud connected private line network and vpc Dedicated line gateway to achieve high availability ; Usually , The cloud networking dedicated line gateway is mainly used ,VPC Supplemented by dedicated line gateway .

Hybrid cloud network disaster recovery solution

Fault deduction ：

1） Some special line channels are abnormal , The traffic will be automatically dispatched to other dedicated lines , Limited impact on business perception .

2） All special line channels are abnormal , The service traffic shall be dispatched to the public network to recover the service , You need to call API Interface disabled VPC Type dedicated line gateway routing , And add VPN/GRE Route to manually resume business . Mainly because of the current VPC At present, type a dedicated line gateway only supports static routing , As a result, the route cannot converge automatically .

3） The cloud networking dedicated line gateway is abnormal , Business traffic will be automatically scheduled to VPC Type dedicated line gateway to restore service , Limited impact on business perception .