基于键值存储的RTB广告买方平台的研究与设计
发布时间：2018-12-29 12:28
【摘要】：随着互联网用户规模的增加,互联网广告行业也在迅速发展。近几年出现的广告实时竞价的模式与传统广告投放模式有很大区别。RTB广告(实时竞价广告)使广告投放的重心转移到受众身上,通过对用户数据的搜集和用户属性的挖掘,来实现广告投放效果的最大化。RTB广告近几年在美国发展迅速,并于2012年由谷歌公司引入中国。 DSP是RTB广告投放系统中的买方平台。DSP为广告商提供管理广告的功能,同时DSP搜集互联网受众的信息,从大量受众数据中分析出受众的兴趣、购物倾向等属性,再根据这些属性进行针对性的广告投放,以提升广告效果。根据受众属性和兴趣投放广告,是DSP业务的核心竞争力。其中用于存放受众数据的存储系统必须有良好的可伸缩性,传统关系数据库在可伸缩性方面很难满足DSP的要求,所以设计合理的并且拥有良好的可伸缩性的大数据存储方案显得尤为重要。 DSP系统中使用键值存储策略来存放受众数据。DSP系统中的受众数据在广告投放过程中将不断的增多,因此受众数据的存储系统必须拥有良好的可扩展性,同时还要有良好的查询性能。本文研究了目前存在的一些常见的键值数据库,设计了适合本DSP系统的键值存储方案。键值存储的关键问题是数据划分、数据备份和负载均衡。本文使用改进的一致性哈希算法解决数据划分问题。引入虚拟节点的概念,数据主键与物理存储节点的映射过程分为两个步骤,第一步将数据主键通过一致性哈希算法映射到虚拟节点,第二步以虚拟节点为单位完成其与物理数据节点的映射。传统的一致性哈希算法难以实现动态负载均衡,本文的负载均衡算法是以虚拟节点为单位进行的,使动态负载均衡的实现成为可能。 DSP系统的主要用户是广告商和广告代理商。广告商上传广告到DSP系统,为广告定制投放定向和投放计划,DSP系统根据广告投放计划为用户投放广告。DSP为广告商提供了地域定向、语言定向和设备定向等七种定向,并综合多种因素对广告请求进行出价。DSP系统的前端界面使用了三级菜单,将系统功能充分展示在用户面前,为用户提供友好的使用界面。在架构上尽量减少网页的深度,使开发逻辑清晰。DSP使用多个广告竞价服务器处理竞价请求,利用web服务器将高频率的竞价请求分散到各个竞价服务器,来达到应用级别的负载均衡。
[Abstract]:With the increase in the size of Internet users, the Internet advertising industry is also developing rapidly. In recent years, the pattern of real-time advertising bidding has been very different from the traditional advertising mode. RTB advertising (real-time bidding advertising) makes the focus of advertising to the audience, through the collection of user data and the mining of user attributes. RTB has grown rapidly in the United States in recent years and was introduced to China by Google in 2012. DSP is the buyer's platform in the RTB advertising delivery system. DSP provides advertisers with the function of managing advertising. At the same time, DSP collects the information of the Internet audience, analyzes the audience's interest, shopping tendency and other attributes from a large number of audience data. Then according to these attributes targeted advertising, in order to improve the advertising effect. According to the audience attributes and interest in advertising, is the core competitiveness of DSP business. The storage system used to store the audience data must have good scalability, and the traditional relational database is difficult to meet the requirements of DSP in scalability. Therefore, the design of reasonable and have good scalability big data storage scheme is particularly important. In DSP system, the key storage strategy is used to store the audience data. The audience data in the DSP system will increase continuously in the process of advertising, so the storage system of the audience data must have good expansibility. At the same time also have good query performance. In this paper, some common key and value databases are studied, and the key and value storage scheme suitable for this DSP system is designed. Key issues in key storage are data partitioning, data backup and load balancing. In this paper, the improved consistency hash algorithm is used to solve the data partition problem. The concept of virtual node is introduced. The mapping process of data primary key and physical storage node is divided into two steps. The first step maps the data primary key to virtual node through the consistent hash algorithm. The second step is to map the virtual node to the physical data node. The traditional uniform hash algorithm is difficult to realize dynamic load balancing. The load balancing algorithm in this paper is based on virtual nodes, which makes the realization of dynamic load balancing possible. The main users of the DSP system are advertisers and advertising agencies. Advertisers upload ads to the DSP system, and the DSP system provides users with advertisements according to the advertising plan. DSP provides advertisers with seven kinds of orientation, such as regional orientation, language orientation and device orientation. The front-end interface of the DSP system uses a three-level menu to fully display the functions of the system in front of the user and provides a friendly interface for the user. DSP uses multiple ad auction servers to process bidding requests, and uses web server to spread high frequency bidding requests to each bidding server. To achieve application-level load balancing.
【学位授予单位】：华中师范大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP311.13;TP393.09
本文编号：2394831
[Abstract]:With the increase in the size of Internet users, the Internet advertising industry is also developing rapidly. In recent years, the pattern of real-time advertising bidding has been very different from the traditional advertising mode. RTB advertising (real-time bidding advertising) makes the focus of advertising to the audience, through the collection of user data and the mining of user attributes. RTB has grown rapidly in the United States in recent years and was introduced to China by Google in 2012. DSP is the buyer's platform in the RTB advertising delivery system. DSP provides advertisers with the function of managing advertising. At the same time, DSP collects the information of the Internet audience, analyzes the audience's interest, shopping tendency and other attributes from a large number of audience data. Then according to these attributes targeted advertising, in order to improve the advertising effect. According to the audience attributes and interest in advertising, is the core competitiveness of DSP business. The storage system used to store the audience data must have good scalability, and the traditional relational database is difficult to meet the requirements of DSP in scalability. Therefore, the design of reasonable and have good scalability big data storage scheme is particularly important. In DSP system, the key storage strategy is used to store the audience data. The audience data in the DSP system will increase continuously in the process of advertising, so the storage system of the audience data must have good expansibility. At the same time also have good query performance. In this paper, some common key and value databases are studied, and the key and value storage scheme suitable for this DSP system is designed. Key issues in key storage are data partitioning, data backup and load balancing. In this paper, the improved consistency hash algorithm is used to solve the data partition problem. The concept of virtual node is introduced. The mapping process of data primary key and physical storage node is divided into two steps. The first step maps the data primary key to virtual node through the consistent hash algorithm. The second step is to map the virtual node to the physical data node. The traditional uniform hash algorithm is difficult to realize dynamic load balancing. The load balancing algorithm in this paper is based on virtual nodes, which makes the realization of dynamic load balancing possible. The main users of the DSP system are advertisers and advertising agencies. Advertisers upload ads to the DSP system, and the DSP system provides users with advertisements according to the advertising plan. DSP provides advertisers with seven kinds of orientation, such as regional orientation, language orientation and device orientation. The front-end interface of the DSP system uses a three-level menu to fully display the functions of the system in front of the user and provides a friendly interface for the user. DSP uses multiple ad auction servers to process bidding requests, and uses web server to spread high frequency bidding requests to each bidding server. To achieve application-level load balancing.
【学位授予单位】：华中师范大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP311.13;TP393.09
【参考文献】
相关期刊论文 前2条
1 ;实时竞价广告(RTB,Real-Time bidding)市场规模将占美国展示广告市场的13%[J];广告大观(理论版);2012年06期
2 胡超晔;张忠能;;Key-value数据库在时序数据及互动信息处理中的分析应用[J];微型电脑应用;2012年11期
相关硕士学位论文 前6条
1 孙皓;基于Erlang和异构平台的Key-Value数据库的研究[D];成都理工大学;2011年
2 霍志刚;中国互联网广告的发展[D];对外经济贸易大学;2007年
3 王习特;Key-Value数据库上连接操作的查询处理与优化[D];东北大学;2011年
4 赵雪霞;基于DHT的key-value存储的范围查询技术研究[D];西安电子科技大学;2011年
5 赵见;高性能高可用键值存储系统的设计与实现[D];电子科技大学;2010年
6 杜磊;基于P2P的key-value存储系统关键技术研究[D];西安电子科技大学;2009年
，本文编号：2394831
本文链接：https://www.wllwen.com/wenyilunwen/guanggaoshejilunwen/2394831.html
