Computing architecture of microblog comments

User behavior modeling and performance estimation

background

Reference resources 《 Microblogging 2020 User development report 》,2020.9 Monthly Weibo users' monthly life 5.11 Billion , Diurnal activity 2.24 Billion . This paper mainly discusses the architecture design of the comment module of microblog .

Suppose that the average person sends 1 Micro-blog ( Only consider text microblog ), The daily sending volume of microblog is about 2.5 Billion bars . Most people tweet in the morning 8:00~9:00 spot , At noon, 12:00~13:00, evening 20:00~22:00, Suppose that the proportion of the total amount of microblogging in these time periods is 60%.

Because the vast majority of microblog users watch microblogs for big V And stars , Therefore, we assume that the average number of people watching a microblog is 100 Time , The number of times you watch microblog is : 2.5 Billion  * 100 = 250 Billion .

behavior analysis

• Most people's time period of watching microblog basically coincides with that of sending microblog , Writing and reading comments are behaviors during the period of reading microblog , Therefore, the time distribution in the background is also suitable for the comment module .

• Suppose you have every look 10 Micro blog , Users will view 8 Microblog comments .

• Suppose that every time a user watches 10 Comments on Weibo , Will the new 1 Your own comments .

Performance estimation

• See the comments ： The number of reviews viewed is ：250 Billion  * 80% = 200 Billion

So during this time , On average, I read the comments QPS The calculation is as follows ：200 Billion  * 60% / (4*3600) = 800K/s.

• Write a comment ： The number of comments is ：250 Billion  * 10% = 25 Billion

So during this time , The average person who writes a comment TPS The calculation is as follows ：25 Billion  * 60% / (4*3600) = 100K/s

High performance computing architecture

The comment system is also a typical scenario of reading more and writing less , To improve usability . Therefore, the service of writing comments and reading comments should be split to ensure independence .

See the comments

Business characteristic analysis

Watching Weibo is a typical reading scene , Because the microblog cannot be modified after it is posted , Therefore, it is very suitable to use cache architecture , At the same time, due to the large number of requests , Load balancing architecture also needs .

Architecture analysis

1.  Over 100 million users , You should use a multi-level load balancing architecture ;2.  The requested quantity reaches 200 Billion , You should use a multi-level cache architecture , In especial  CDN  cache , Is the core of cache design .

Architecture design

1. If the load balancing algorithm is selected, visitors can directly watch the microblog , Therefore, sending a request to any server can , Choose here “ polling ” perhaps “ Random ” Algorithm .

2.  Business server quantity estimation assumptions  CDN  Capable of carrying 90% Of users , So the rest 10% The request to read microblog enters the system , Ask for  QPS  by 800K/s * 10% = 80K/s, Because the processing logic of reading microblog is relatively simple , Mainly read cache system , Therefore, it is assumed that the processing capacity of a single business server is 1000/s, Then the number of machines is 80 platform , According to a certain reserved amount , The final number of machines is 100 platform .

Write a comment

Business characteristic analysis

Microblogging is a typical writing operation , So you can't use cache , Load balancing can be used .

Architecture analysis

Over 100 million users , You should use a multi-level load balancing architecture , Cover  DNS -> F5 -> Nginx ->  Multilevel load balancing of gateway .

Architecture design

1.  When the load balancing algorithm chooses to send microblog, it depends on the login status , The login status is generally saved in the distributed cache , So when you tweet , Send the request to any server , Choose here “ polling ” perhaps “ Random ” Algorithm .

2.  The number of business servers is estimated. Microblogging involves several key processes : Content review ( Rely on the audit system )、 Write data to storage ( Dependent on storage system )、 Data write cache ( Rely on the cache system ), So, per service per second 500 To estimate , complete 100K/s Of  TPS, need 200 Servers , Plus a certain amount of reserve , The final number of machines is 240 platform .

Overall architecture scheme

High availability computing architecture for hot events

For sudden hot events , It is characterized by unpredictability , Instant reading and writing comment traffic is large .

• View comments for , All read requests . It can be monitored through operation and maintenance , When the load of the cluster is detected to surge in a sliding window , Or, after the load of the entire cluster reaches the threshold, it can dynamically expand the capacity to add machines and service instances .

• For writing comments , Leakage bucket algorithm with current limiting , Cache the user's write request first , When using background threads to write to storage .