• Definition
  • Defining the architecture, components, modules, interfaces and data for a system to satisfy specified requirements
• Conceptual design -> Logical design -> Physical design (Macro -> Micro, can also Micro -> Macro)
• What is a good design?
  • Healthiness
    • Execution
    • Communication
  • Simplicity
    • No more, no less
    • Understandable

• Focus in this lesson: Fundamental questions in system design interviews
  • Design the system
  • Evaluate query per second
  • Scale the system
• Design Netflix/ytb/Spotify
  • Tag:Uber, Google, Alibaba

Please design "Netflix" - Macro

• Crack a design in 5 steps:
  • Scenario: case/interface
  • Necessity: constrain/hypothesis
  • Application:service/algorithm
  • Kilobit:data
  • Evolve
• Scenario:case/interface
  • 1. Enumerate (chat w/ interviewer)
       1. register/login
       2. play movie
       3. movie recommendation
       4. ...
    2. Sort
       1. play movie
          1. get channels
          2. get movies in channels
          3. play a movie in a channel
       2. ...
• Necessity: constrain/hypothesis（主要经验估算，辅以询问数据）
  • 1. Ask: how many active users?
    2. Predict
       1. User
          1. average concurrent users = $$\frac{daily-active-users}{daily-seconds} * average-online-time$$
          2. peak users = average concurrent users * 6 (6是经验值)
          3. MAX peak users in 3 months = peak users * 2 （上线后三个月可能用户增长）
    3. Traffic
       1. Traffic per user = 3Mbps
       2. Max peak traffic = Max peak users * traffic per user
    4. Memory
       1. Memory per user = 10KB
       2. Max daily memory = daily active users * 2 (3 months) * 10KB = 100GB(Redis TB OK)
    5. Storage
       1. Total # of movies = 14,000
       2. Movie storage = # of movies * average movie size(multiple versions) = 14,000*50GB = 700TB
• Application:service algorithm
  • 1. Replay the case, add a service for each request
    2. Merge the service

• Kilobit: data
  • 1. Append dataset for each request below a service
    2. Choose storage types
       1. User Service - Accounts(MySQL)
       2. Channel Service - Channel List(MongeDB?? -> 档案)
       3. Movie Service - Movies(Files)
• Evolve
  • 1. Analyze
       1. Better: constraints
       2. Broader: new cases
       3. Deeper:details
    2. Views..
       1. Performance
       2. Scalability (# of users.. # of machines)
       3. Robustness

Please design "Netflix" - Micro

• Design recommendation module

• 10^6~10^9 approximately 1s
• Similarity <- bucket sort (倒序索引 Inverted Index，以movie为key建立索引)
• Dispatcher -> load balancer

• 密集型：
  • 硬盘密集 - 读写？，爬虫结果保存
  • CPU密集 - 计算密集，爬虫内部去重
  • 内存密集 - 爬虫结果二次计算
  • 网络带宽密集 - 爬虫爬取网页

Improve Robustness

• Dispatcher 可能有mirror
• Feed manager -> timeline manager 信息流
• Loggers -> Monitor