（GeekBand）系统设计与实践 案例分析

案例

• News Feeds
• Stats Server
• Web Crawler
• Amazon Product Page

News feed（信息流）

Define feed

Organize

• aggregate（分类）
• dedup（去重）
• sort（排序）

Level1.0

Database Schema:

• User
• Friendship
• News

GetNewsfeed:

• merge news
• Newsfeed vs News

Why bad?

100+ friends

1Query-->Get friends list

1Query-->

SELECT news

WHERE timestamp>xxx
AND sourceid IN friend list
LIMIT 1000

IN is slow

Either Sequential scan or 100+ index queries

Level 2.0

Pull vs Push

Pull:Get news from each friend,merge them together.(NewsFeed generated when user request)

Push:NewsFeed generated when news generated.(we have another table to store newsfeed,may cause duplicate news)

Push:

1Query to select latest 1000 newsfeed.
100+ insert queries(Async)

Disadvantage:News Delay.

Level 3.0

Popular star(Justin Bieber)

Flowers 13M+

Async Push may cause over 30 minutes(13M+ insertions,delay too long)

Push+Pull

for popular star,don't push news to flowers

for every newfeed reqiest,merge non-popular user newfeed(push) and popular users newsfeed(pull)

Level 4.0

Push disadvantage

• Realtime
• Storage(Duplicate)
• Edit

Go back to PULL:

• Cache users' latest (14days) news
• Broadcast multiple request to multiple servers(Shard by userld).
• Merge & sort newsfeed
• Cache newsfeeds for this user with timestamp

Click Stats Server

How are click stats stored

A poor candidate will suggest write-back to a data store on every click

A good candidate will suggest some form of aggregation tier that accepts clickstream data,aggregates it,and writes back a persistent data store periodically

A great candidate will suggest alow-latecy messaging system to bugger the click data and transfer it to the aggregation tier.

If daily,storing in hdfs and running map/reduce jobs to compute stats is a reasonable approach

If near real-time,the aggregation logic should compute stats

PS：要如何统计鼠标点击的次数以及相关区域呢？普通的程序员会将每次点击的数据（log）直接存储在数据库一层。比较好的程序员会在前段与数据库间加一个中间层，为点击的数据流做一次聚合，每隔一段时间（1分钟或10分钟）做一次刷新，存储到数据库，大大减轻了后端的压力。优秀的程序员综合以上的两种情况，对于数据量很大，实时性效果不高的情况下，可以通过分布式的批处理方式，将刷新聚合层的时间定位一天。对于时效性强的要适当缩短刷新时间。

Cache Requirement

• When a request comes look it up in the cache and if it hits then return the response from here and do not pass the request to the system.
• If the request is not found in the cache then pass it on to the system.
• Since cache can only store the last n requests,Insert the n+1th request in the cache and delete one of the older requests from the cache
• Design one cache such that all operations can be done in O(1)-lookup,delete and insert.

PS:如何设计cache（LRU设计相关）：

• 在层中缓存部分请求的处理方式，如果接收的请求在层中存在对应的处理方式，则无需把请求发送到后端系统
• 如果在层中找不到对应处理，则发送需求到后端
• 以定长队列的形式缓存，缓存最近的n个需求，头进尾出
• 将层中的匹配操作算法控制在O(1)范围

Web Crawler

爬虫

Amazon Product Page

The product page includes information such as

• product information
• user information
• recommended products(what do other customers buy after viewing this item,recommendations for you like this product,etc)

Reference

• http://highscalability.com
• The Log:What every software engineer should know about real-time data's unifying abstraction
• Job Interviews:How should I prepare system design questions for Goole/Facebook Interview?
• HOW TO ACE A SYSTEMS DESIGN INTERVIEW
• http://www.hiredintech.com/app