Recently, the famous blogging community Tumblr was acquired by Yahoo for $1.1 billion. Why Yahoo attaches so much importance to Tumblr, the following is an introduction to the technical architecture of Tumblr.
I planned to translate it myself, but I found that someone has already done so, so I reprinted it directly.
===========================================================================================
In the recent news we learned that Yahoo bought Tumblr for $1.1 billion: Yahoo bought Tumblr for $1.1 billion . You may find the introduction that Instagram was also bought heavily by Facebook . Is this a coincidence? That's up to you
Why did Yahoo buy Tumblr? I may not be able to judge the commercial value of this deal, but if you know the technical aspects of Tumblr, you will definitely give Tumblr a thumbs up. Why do you say that, please read on.. .
With more than 15 billion page views per month, Tumblr has become a popular blogging community. Users may like it for its simplicity, beauty, and its strong focus on user experience, or its friendly and busy way of communicating, but in short, it is loved by people.
Growth of more than 30% per month is certainly not possible without challenges, and reliability issues are particularly daunting. 500 million pageviews per day, 40,000 requests per second peak, 3TB of new data storage per day, and running on more than 1,000 servers, all of which help Tumblr achieve massive scale.
For startups to succeed, they have to cross the threshold of a dangerous period of rapid growth. Looking for talent, constantly revamping the infrastructure, maintaining the old architecture, and facing a huge increase in traffic every month, and there used to be only 4 engineers. That means having to make tough choices about what to do and what not to do. That's the state of Tumblr. Fortunately, there are now 20 engineers and can have the energy to solve problems and develop some interesting solutions.
Tumblr started out as a very typical LAMP application. Currently evolving to a distributed service model based on Scala , HBase , Redis , Kafka , Finagle , in addition to an interesting Cell-based architecture for Dashboard support. The focus is now on solving their PHP programs short-term problems, identify problems, and properly use servicization to solve them. The biggest problem with Tumblr right now is how to transform it into a large-scale website. The system architecture is evolving from LAMP to a state-of-the-art technology portfolio, and the team is also evolving from a small entrepreneurial type to a well-armed, on-call regular development team, constantly creating new functions and infrastructure. The following is Blake's introduction to the Tumblr system architecture. Tumblr URL: http://www.tumblr.com/ |
statistics
software
硬件
|
构架
|
老的Tumblr构架
|
新的构架
|
内部通讯管道(Firehose)
|
为 Dashboard 收件箱设计的 Cell 架构
|
在纽约启动运作
团队架构
|
软件部署
|
开发
招聘流程
|
经验及教训
- 自动化无处不在
- MySQL(增加分片)规模,应用程序暂时还不行
- Redis 总能带给人惊喜
- 基于 Scala 语言的应用执行效率是出色的
- 废弃项目——当你不确定将如何工作时
- 不顾用在他们发展经历中没经历过技术挑战的人,聘用有技术实力的人是因为他们能适合你的团队以及工作。
- 选择正确的软件集合将会帮助你找到你需要的人
- 建立团队的技能
- 阅读文档和博客文章。
- 多与同行交流,可以接触一些领域中经验丰富的人,例如与在 Facebook、Twitter、LinkedIn 的工程师多交流,从他们身上可以学到很多
- 对技术要循序渐进,在正式投入使用之前他们煞费苦心的学习 HBase 和 Redis。同时在试点项目中使用或将其控制在有限损害范围之内。
.....
译文地址:http://www.linuxeden.com/html/news/20130526/139587.html