8 million visitors per day, 5 9 availability architectures of 0.2 million RPS websites

The number of visits per day reaches 8 million, and the number of requests per second reaches 0.2 million. The popularity of AOL is indeed amazing. However, AOL has always maintained 99.999% availability under such high traffic. Here we go into the architecture designed six years ago.

AOL.com, American Online, a famous news aggregation website in the United States, is also one of the largest Internet service providers in the United States. Currently, the number of daily visits to AOL.com exceeds 8 million, and the monthly PV reaches 1 billion. Recently, Dave Hagler, founder of Lead Squirrel, AOL architect and engineering team leader, and having 10 years of Architecture Experience, wrote an article on HighScalability to share the AOL high availability (99.999%) website architecture:

Currently, the AOL.com architecture has evolved to the fifth generation, which can be said to have been rebuilt five times in 20 years. Currently, the architecture used was designed six years ago, although the overall architecture remains unchanged, however, the update and addition of components have never been stopped. During the six-year continuous improvement process, the code, tools, development, deployment, and other aspects were fully optimized, and AOL.com stood down in the current data flood.

The AOL.com engineering team has been around 25 people, including developers, testers, and O & M personnel. The company mainly works in Dulles and Virginia, and a small part in Dublin and Ireland. It is never easy to deal with such traffic and keep the availability at five or nine. In the system architecture, we use a variety of technologies: java, JavaServer Pages, Tomcat, Apache, CentOS, Git, Jenkins, Selenium, and jQuery.

Design principles

We have a clear idea about the architecture design. The most important thing is redundancy. When a part of the system fails or requires offline maintenance, it is no doubt that a backup will save time and effort, the availability of the 5-9 protocol requires no more than 5 minutes of downtime each year.

The second principle is that AOL.com cannot rely on any shared infrastructure to deliver pages. Even if a system or content fails, AOL.com still needs to maintain high availability. Shortly after the design of this architecture, most of the Network Content of AOL shares an infrastructure called Big Bowl, which has a very fatal problem-the interaction between different content. To solve this problem, the current AOL specifically designed the isolation between different content. Any content dependent on AOL.com will be moved by a protection service to a group of fewer hosts, the protection service is responsible for aggregating calls to downstream systems. Therefore, instead of receiving requests from tens of thousands of servers, the downstream system may only obtain requests from 20 different servers, and the response will also be cached to reduce the load. At the same time, the O & M team will back up the External Database on AOL.com. In this way, only the network and protocol services in the system are shared.

Physical infrastructure

AOL.com is deployed in three different data centers, two in Northern Virginia and one in California, which are independently operated by the company. Although the size of each data center is sufficient to support the entire AOL.com, AOL.com still insists on running the three data centers at the same time. Multi-redundancy makes offline maintenance of the data center easy.

When a request is sent for access, the Akamai GSLB responsible for cross-Data Center Server Load balancer directs the user to the nearest data center. Akamai CDN is used for static content. Once a data center is determined, a request (database or service) is sent to the data center. The user's session information is stored in cookies and sent through the request. Because the request does not need to be saved, the request can be executed on any server.