High performance and high availability computing architecture scheme commented by Weibo

（ One ）  Calculation performance description

•   Estimated number of users

  According to the microblog 2022 The results of the first quarter of , Daily active users reached 2.52 Billion users . That is, the user base is 2.52 Billion

• Behavior modeling

  Users post comments on Weibo information , Most of them are aimed at stars or news information , It is expected that every microblog , The average is about 10 People will have 1 People post comments about . The frequency of users' viewing comments every day will be lower than that of users' viewing microblogs , That is, check the comments every day about 10 People will have 2 About people .

• Performance estimation

  Post comments ： 2.52 Billion  

* 60% /（4 * 3600）* 0.1 *

 

0.8 = 800/S

  View comments ： 2.52 Billion  

* 60% /（4 * 3600）* 0.2 *

 

0.8= 1.6K/S 

  notes ： 80% The comments of occurred after the microblog was posted .

( Two )  analysis

• Functional analysis

• For comments posted by users , The requirement for timeliness is not high , Can cache storage and load balancing policy rates .

•   For users to view comments , You can use the local cache and CDN And other caching methods and load balancing policies .

•   For users, posting and viewing comments are on the same page 【 The same scenario triggers 】, So combine the two functions into one application service , No more splitting .

• Load architecture recommendations

  The number of users is large , A multi-level load balancing architecture should be used , Mainly for DNS -> F5 -> Nginx -> Multi level load balancing architecture of gateway .

• Cache architecture recommendations

  The timeliness of users' comment data is not high , Whether you are posting comments or viewing comments, you can use cache temporary storage .

（ 3、 ... and ） Design

• Load balancing algorithm selection

  Weibo stores the basic information of individual accounts through distributed shared cache , User comment data can be loaded on any server at will , The load balancing algorithm can select random and polling modes .

• Estimate the number of business servers

  The microblog comment service adopts the mode of Multiservice deployment , The process includes   Content review 、 Write cache 、 Data warehousing , Therefore, it is estimated that each server can process 1K/s, According to the budget analysis, the performance index is 800/S[ Post comments ]、1.6K/S[ View comments ],  Therefore, it is estimated that 18 Servers [ There is some redundancy ].

（ Four ） High performance architecture design description

• The load architecture is designed as follows ：

• Most microblog users use mobile phones App[ Microblogging ], establish HttpDNS Server side [ It is also a cluster deployment mode ] Efficient load for applications .

• Establish a four layer load balancing structure , Respectively DNS(HttpDNS) layer 、 Hardware load layer (F5/A10)、 Software load layer （Nginx）、 Application gateway load layer .

Post comments and view the comment load schema

• The multi-level cache design architecture is as follows ：

  Retain some of the original browsing information through the local cache [ Comments are sorted by time ], The content of the comments can be cached and stored in the process to which the application service is applied 【 The timeliness of the business is low 】.

（ 5、 ... and ） High availability architecture design description

• Analysis of hot events ：

  Comment on ： After the hot event , Most of the requests fell on the microblog that led to the hot event . There will be a lot of comments under this microblog content .

• Technical architecture analysis ：

  For the comment writing itself, a caching mechanism has been adopted , There should be no special performance for hot issues . For reading comments , You can use the comment data as zset Data structures are stored in redis in , Store in each copy , In this way, the application can provide high availability when reading . Apply periodic updates later zset Data in .