The China Railway Customer Service Center website (12306.cn) is one of the world's largest real-time trading systems. At the peak of the Spring festival in early 2012, 20 million people visited the site every day, with a maximum of 1.4 billion clicks per day. A massive influx of simultaneous internet access caused 12306 to be nearly paralyzed. To this end, starting in March 2012, the Railway Corporation (formerly the Ministry of Railways) began research, transformation 12306. June 2012 selected Pivotal GemFire distributed Memory Computing platform Transformation 12306, by the Iron Academy and the company Hongyu (IISI) Information Technology Co., Ltd. to provide technical implementation.
Click to see the big picture
First phase of the transformation of the main bottleneck of 12306-------ticket query system. September Complete the Code transformation, the system on-line. 2012 National Day, in the online booking peak period, you can log in 12306, although it is difficult to book tickets, but the query is very fast. October 2012, the second phase of the GemFire transformation of the Order query system (customer inquiries of their own order records). 2013 Spring Festival, in the online booking peak period, you can log in 12306, although it is difficult to book tickets, but the query is very fast, and inquires their own booking and orders are also very soon.
According to the system Operation data record, after the technical transformation, only uses 10 several X86 servers to realize the previous dozens of small machine's remaining votes computation and the inquiry ability, the single query's longest time from the preceding 15 seconds to descend to 0.2 seconds below, shortened 75 times times above. During the 2012 travel season, the extremely high flow of traffic concurrency, support tens of thousands of concurrent queries per second, peak period reached 26,000 QPS throughput, the overall system efficiency significantly improved.
Order Query System transformation, in the system before the transformation of the operating mode, only 300-400 QPS per second to support the throughput, high flow of concurrent queries can only be achieved through the sub-Library. After the transformation, you can achieve up to tens of thousands of QPS throughput, and query speed can be guaranteed in about 20 milliseconds. The new technology architecture can be dynamically expanded on demand, and it can be responded by dynamically increasing the X86 server to maintain the millisecond response time.
The principle of GemFire distributed memory data platform is:
Through cloud Platform virtualization technology, the memory of several X86 servers is pooled to form a pool of memory resources up to dozens of TB, loading all data into memory for memory calculation. The calculation process itself does not require a read-write disk, but periodically writes data synchronously or asynchronously to disk. GemFire stores multiple data in a distributed cluster, any machine fails, backup data on other machines, no worries about data loss, and disk data as a backup. GemFire supports the persistence of memory data into a variety of traditional relational databases, Hadoop libraries, and other file systems. 12306 before the use of UNIX minicomputer architecture, the use of GemFire technology transformation into a LINUX/X86 server cluster architecture.
12306ng (open source project for 12306 sites) There is no comment on this. However, yesterday was the first day of the Spring Festival train ticket sale, it is reported that 12306 website "paralysis" 1 hours, and the "serial number" problem, obviously, the simple distributed memory data technology is not enough to meet the technical challenges of 12306 face. After all, even the Hana of SAP, and the GemFire of pivotal, are new.