Design And Realization Of Key Improvements In Service System Of Large Scale E-commerce Website

As the Internet dive into people’s daily life from all aspects, electronic commerce became more and more popular. According to the related survey data, electronic commerce industry has made a great development, in terms of Gross Merchandise Volume and the number of active user. The number of simultaneously online user became more than 10 million in rush hour. The page view per day of the website maintained as more than 1 billion. Because of the characteristics of the electronic commerce websites, we must guarantee that our users can view the webpages in less than 4 seconds, in order to get a good conversion rate. Also the website must make sure its service is always available in 24 x7 hours. All of these make a strict requirement for the electronic commerce website on its overall design. The usual architecture in Java is using Web framework, Web Service and Database technology. This kind of architecture model cannot satisfy the user request, in the case of massive concurrent access. The mainly problem focus on two aspects. First of all, the efficiency of the web application will decrease sharply and apparently, while it’s handling more than 10000 concurrent connection requests. Secondly, the scalability is very bad, the web tier and the persistence tier have high coupling, and each tier cannot maintains scalability and high availability. The only way to get increment of system capacity is application clone and load balance, which is simple but ineffective.The combination of J2EE technology, distributed cache technology and distributed service technology offer an easy way to design and implement a stock-management distributed service, which has the ability to handle high concurrency and scalability.First, the event-driving mechanism and asynchronization technique are very useful in improving the capacity and performance for large quantities of concurrent accesses.Secondly, after analyzing the full workflow, using the proxy pattern to reform the cache system and database system can eliminate the single point failure risk, and offer the cache system and database system scalability and availability.Finally, using parallelization pattern in service orchestration will make the whole service respond more quickly. Using the long-connection pattern in network data transfer make the invocation between distributed services more effective.Based on the methodologies mentioned above, the whole system implements a clearly layered design, which can be divided into presentation layer, distributed service layer and data persistent layer. Each layer remains high performance, high availability and scalability respectively, and keeps the ability of avoiding the single-point failure and performance bottleneck. Even facing the traffic explode, the system can scale up some specific layer quickly, depending on the metric data of each layer.