Actually, I fantasized about this issue a long time ago. Before that, it was in the process of looking for a person, repeatedly being sent a good person card. , which made me start to wonder if I never met the right person. I fantasize that if the country has an AI system that intelligently assigns objects, it would be good. I believe that many partners in the process of looking for a partner, have suffered from not being able to find the right person. Especially for those who do not love social partners, originally the circle is not big, to find the object is even more difficult. But if one day the state develops a reliable couple matching system through big data, are you willing to give it a try? Hang the DJ from Black Mirror Season 4 tells the story of an AI matchmaker. The story is set in the future sometime in the country in order to reduce the divorce rate less unfortunate marriages, all unmarried men and women will be sent to a certain city for AI testing, and then assigned a marriageable partner. The duration of the trial marriage varies, and after several trial marriages, the system will arrange for the most compatible match to be married.

Technology behind big data love

For example, dating apps are the most common tools for people to find love with big data, and here's a look at the technology used in dating apps

Individual characteristics and user profiles

To find love with big data, the first step is to tell the computer to find love. To find love with big data, the first step is to tell the computer who you are, your age, gender, hobbies, job, income, height, looks, etc. The second step is to tell the computer what you want. The second step is to tell the computer what you want. Specify your criteria, how old you want him/her to be, where you want him/her to be, what kind of job he/she has, how much he/she earns, and possibly a variety of labels - intellectual, cute, short hair, etc. This way, the algorithm can use the information you provide to find love. The algorithm can then use the information you provide to calculate a vector for you, which may be hundreds, thousands or even tens of thousands of dimensions, and in the future, the computer will use this vector to represent you in the algorithm. And the process of doing this is called user profiling. China is a populous country, the number of users is very large, need to store the user profile is also very large, in order to be able to efficiently extract and calculate, the user data are stored in the database, the following figure is Baidu built in-memory database Feed-Cube, and the use of the KV (Key-Value, key-value pairs) of the storage structure, which can provide a very good read and write performance.

But with the increase of users, the memory slowly began to store so much user profile data, once the memory overflow (Out of Memory) will be allowed to machine, causing serious problems. A simple solution is to add memory, a single consumer-grade memory ceiling is 32g, server-grade memory has 64GB, 128GB rare and expensive. Adding memory is either not possible or too expensive, is there a better way? Of course, there is a better way to do this, and for this reason, Intel has proposed the "Ardent" persistent memory technology. Ascendant Persistent Memory combines the advantages of existing DRAM memory and SSDs without their obvious shortcomings. With its innovative media, Ascendant Persistent Memory provides read/write performance and latency close to that of traditional memory, storage capacity closer to that of SSDs, and support for data persistence and higher price/capacity than DRAM memory. persistence and higher price-to-capacity ratios than DRAM memory, and durability unmatched by SSDs. This makes it ideal for multi-user, high concurrency, and high capacity scenarios, and is especially suited for scaling memory to carry larger volumes of data that require higher read/write speeds and latency to process.

For example, this type of recommender system is currently widely used in the field of small video platforms such as a sound. It must be said that the recommendation algorithm of a certain sound is really remarkable. Due to the precise recommendation, many of our friends will be addicted to it as soon as they open a certain sound. The videos recommended by the system are too much to their taste. The back of a certain sound recommendation system relies on the self-developed volcano engine. But with the continuous expansion of the business, the volcano engine, the recommended system of data storage and processing needs are also increasing day by day, coupled with Internet users on the latency and relevance of the content recommended by the increasingly demanding requirements, the volcano engine began to IndexService applied to its recommended business, in order to provide high-level syntax tree + complex merged tree + inverted indexing capabilities, and to achieve KV query capabilities.

However, the demand for access to data brought about by these initiatives has further increased the performance requirements for IndexService, which in turn has placed higher demands on the storage server's throughput, responsiveness, stability, and ability to recover from unexpected interruptions. Volcano Engine has made a major breakthrough in the field of memory-storage architecture by introducing Intel@Armor Persistent Memory and combining it with Intel@Xeon@Scalable processors to y optimize the IndexService storage architecture. In order to meet the storage capacity required for recommender system IndexService to support large-scale access, such as a service it carries that can reach a total TPS of 70 million reads and 30 million writes during the evening peak period, and to enhance its data access and processing capabilities, the Volcano Engine configures Intel@Armor Persistent Memory in App Direct mode when using it, realizing that applications can access data on persistent memory directly from the user's The App Direct mode enables applications to directly access data on persistent memory from user space, eliminating the need for additional overhead and meeting the high data access speed requirements of users.

Data security is critical

After a dating app stores a large amount of user data, the security of user data is critical. When big data has specific information about each person, security is an issue. So much sensitive information is stored in the cloud platform. Once leaked, then everyone's privacy is nakedly exposed in the hands of lawless elements. In recent years, a variety of pig-killing disks have been frequently exposed, leaving many people unable to defend themselves. The real world is still difficult to distinguish between the real and the fake, then the network is with easy to make a fake. How can these cloud platforms or databases guarantee the security of these data? Intel SGX I think is a good solution to these potential problems. Intel@Software Guard Extensions (Intel@SGX)12 provides hardware-based memory encryption that isolates application-specific code and data in memory. This technology is primarily about securing data with new controls for cloud and enterprise environments.

The current generation of Intel@Xeon@ Scalable processors feature enhanced Intel@Software Guard Extensions (Intel@SGX) technology, which allows applications, to be divided into processor-enhanced compartments that enhance the protection of specific applications or data from disclosure or modification. Before SGX was used, database data was directly exposed to system applications. However, with SGX, the database data is isolated and protected in an area for use by other programs. This way our sensitive data will not be misused by some undesirable programs.