朱金生-新型云架构提升下一代移动应用和可穿戴设备用户体验


A new cloud Infrastructure powering the next-generation mobile experiences 新型云架构提升下一代移动用户体验 George Chu 朱金生 Distinguished Architect 雅虎北京全球研发中心副总裁及首席架构师 10.17.2014 Yahoo today Yahoo! Proprietary and Confidential #1 GLOBALLY >800M MAU 活跃用户 Y! SPORTS 105M MAU 活跃用户 Y! FINANCE 76M MAU 活跃用户 MOBILE 430M MAU 活跃用户 Y! NEWS >200M MAU 活跃用户 Others… 2 Science-Driven Engineering 科学驱动的工程开发 Cloud Computing 云计算 Yahoo Beijing Global R&D Center 雅虎北京 全球研发中心 Yahoo today – Powered by Yahoo Beijing Yahoo! Proprietary and Confidential 3 #1 GLOBALLY >800M MAU 活跃用户 Y! SPORTS 105M MAU 活跃用户 Y! FINANCE 76M MAU 活跃用户 MOBILE 430M MAU 活跃用户 Y! NEWS >200M MAU 活跃用户 Others… Mobile Platform 移动平台 Personalization 个性化 Advertising 广告 Internet of Things & Everything (万)物联网 Yahoo! Proprietary and Confidential 4 “Fixed” Computing 固定 (you go to the device) Mobility 移动 (the device goes with you) Internet of Things 物联网 (age of devices) Internet of Everything 万物联网 (people, process, data, things) Year 1995 Year 2000 Year 2013 Year 2020 200M 10B 50B Source: Cisco – Growth of the number of things connected to the internet. Challenges ahead 未来的挑战 . From web to mobile 从传统互联网到移动应用 . Dwell time per article less on mobile than web 移动用户的驻留时间在每篇文章上有着明显的减少 . Notifications to re-engage users 重新参与用户的消息推送 . Interactions between multiple devices 多设备间的交互 . Wearables collecting data more personal and in bigger volume 来自于可穿戴设备的大数据 . Security, privacy and trustworthiness more important than ever 安全,隐私,诚信将变得更加重要 Yahoo! Proprietary and Confidential 5 Source: Yahoo Lab – Relationship between binned average dwell time and article length (per device type). Opportunities abound 机会也很多 Yahoo! Proprietary and Confidential 6 An ecosystem of connected devices whose seamless experience requires a design approach based on Consistency, Continuity and Collaboration principles. 一个生态系统的连接设备,其无缝的体验需要基于一致性、连续性和合作原则的设计方法。 Web Tablet Phone Consistency Continuity Collaboration Web Tablet Phone Consistency Continuity Collaboration Opportunities abound 机会也很多 Yahoo! Proprietary and Confidential 7 An ecosystem of connected devices whose seamless experience requires a design approach based on Consistency, Continuity and Collaboration principles. 一个生态系统的连接设备,其无缝的体验需要基于一致性、连续性和合作原则的设计方法。 New cloud 新云端 Yahoo! Proprietary and Confidential 8 A new cloud infrastructure that supports Internet of Things and that realizes Consistent, Continuous and Collaborative (3C) user experiences. 一个崭新的云基础设施来支持物联网和实现有一致性、连续性、有合作性的用户体验。 Challenge Requirement Architectural Vision From web to mobile 从传统互联网到移动应用 A sustainable business model 一个可持续的商业模式 Accommodate O2O, B2C, B2B2C models 适应O2O、B2C、B2B2C模式 Dwell time less on mobile 移动用户的驻留时间明显的减少 Personalize everything 完全个性化 One unified data pipeline processing 统一管道的数据处理 Notifications to re-engage users 重新参与用户的消息推送 24/7 reliable services, anywhere 提供7*24无间断的服务 Effective pull, stream & push 有效的拉、流、推送 Interactions between multiple devices 多设备间的交互 3C-integrated experiences across devices 有一致性、连续性、合作性的跨设备用户 体验 Single version of truth on data & analytics 真实版本的数据和分析 Wearables collecting more data 来自于可穿戴设备的大数据 Connect people, process, data & devices 连接人、流程、数据和设备 Support new signal types 支持新的信号类型 Security, privacy, trustworthiness problems 安全,隐私,诚信问题 Security, privacy & trustworthiness are number one design goal 安全、隐私、诚信为首要设计目标 Confidently and timely exchange & share data across devices 自信并及时交换和共享数据 INTRODUCING YAHOO’S EXISTING MOBILE & CLOUD ASSETS 引入雅虎现有的移动和云资产 Hadoop – Incredible scale at Yahoo Yahoo! Proprietary and Confidential 10 800 Developers 开发人员 17.2M Jobs/Month 作业/月 460 Products 产品 Hadoop集群提升全球的用户体验 35,0001 Servers 服务器 440 Petabytes Storage 存储空间 10M Compute Hours/Day 计算小时/天 世界上最大的Hadoop集群 1 Across Hadoop, HBase, Storm, Data Loading, and Dev Clusters. Hadoop – Behind every click at Yahoo Yahoo! Proprietary and Confidential 11 CONTENT PERSONALIZATION 内容个性化 AD TARGETING 用户定标 处理超过10亿的网页来提升搜索体验 ANTI-SPAM & ANTI-ABUSE 反垃圾/作弊 每天以大于95%的置信度从200万注册信 息中过滤掉25%的虚假注册 单独基于集群的方案让发布新站点的时 间从几个季度缩短到几个星期 CONTENT AGILITY 内容机动存取 每天预处理1千亿事件(35TB)的用户活动 记录以供下游应用分析 DATA PIPELINES 数据流水线 同时为100万个以上的合约提供广告投放 预测和决策 AD OPTIMIZATION 广告优化 更相关的内容意味着更好的广告效果 Notification – Industry-wide problems END-USER PAIN POINTS . Dropped messages 没有成功推送的消息 . Duplicated messages (i.e. on one device) 推送重复的消息 . Spamming messages 垃圾消息推送 . Untimely-sent messages 不及时或不适宜时间的消息推送 . Replicated messages (i.e. on multiple devices) 在不同设备上推送重复的消息 . Non-engaging messages 用户参与度不高的消息推送 Yahoo! Proprietary and Confidential 12 0 10 20 30 40 50 60 70 80 90 Ride Sharing 打车服务 Food & Beverage 饮食 Financial Services 财经 Betting 赌博 Sports 运动 Online Services 在线服务 eCommerce 电子商务 Travel 旅行 News & Media 新闻及媒体 Social 社交 NOTIFICATION ENABLE RATES PER APP CATEGORY ON iOS PLATFORM1 1 Source: Kuhuna Mobile Analytics. OnePush – Yahoo’s notification standard Yahoo! Proprietary and Confidential 13 First to engage 抢先与用户进行参与 Higher notification opt-ins 提高用户接收推送率 OnePush – Logical architecture Yahoo! Proprietary and Confidential 14 OnePush is a cloud service providing intelligent, personalized and cost-effective 1:1 and 1:Many (in Millions) messaging that span multiple channels and scale to hundreds-of-millions of devices on mobile and web. OnePush Server - to - Server(s) Reliable Messaging Subscription State Attribute Processing API Layer Routing Layer Delivery Layer Registration State Notification State OnePush Connection Serving Streamable Non-Streamable HTTPS SSL APNs GCM Email SMS Streaming Patterns Notification Patterns Schedule a delay- sent notification. Narrowcast to a user or a target. Multicast to a topic subscribed by many. Unicast to a user or a target. Notify Stream Register Stream Notify Subscribe Deliver a message based on a specified condition. Broadcast to live connections. Flurry – Personalizing the mobile experience Yahoo! Proprietary and Confidential 15 Flurry’s mission is to optimize the mobile experience through better apps and more personal ads. FROM WEB TO MOBILE . New Behavior 新用户行为 – from Content Consumption, to Content Sharing, and now to Interactivity, Productivity and Personal Health & Fitness . New Persona 新人物角色 – App Developers (in addition to Users, Advertisers, Publishers and Partners) . Push Model 推送模型 – New user engagement (in addition to Pull and Stream models) . O2O (Online to Offline) 线上到线下商业模式 – New enabled business model (in addition to B2C, B2B2C, etc.) BETTER APPS THROUGH MOBILE ANALYTICS Number of apps Flurry sees 看到应用的数量 – 540,000 Number of app sessions per day 每日应用段量 – 5.5 Billions Mobile Services – Cloud for mobile . Device Catalog 设备检测 – Currently up to 99.7% detection coverage for device, browser, app and network types . Smart Photo Rescaling 智能图像缩放 – Bandwidth-sensitive, form factor appropriate, content-relevant, and GPGPU- powered . Cloud KV Key-Value云存储 – Multi-device app cookie store . Cloud Repository 云知识存放库 – Mappings that relate users, devices, apps and activities . Cloud Object Store 非结构化数据云存储 – Unstructured data at multi-petabyte scale Yahoo! Proprietary and Confidential 16 Web 网络 Tablet 平板 Phone 手机 A set of cloud services collectively optimize for cross-app, multi-devices user experiences. Wearable 穿戴式 Personalization – Know what users want Yahoo! Proprietary and Confidential 17 Personalized content recommendation for Yahoo’s Home Page on both web and mobile. . Content Pipelining 内容处理 – Support streaming & batching from ingesting [content & signals], to content reprocessing [CAP], to topic modeling, to semantic analysis, and to feature extraction . User Profiling 用户建模 – Long-term (batching) and short-term (streaming) user behavioral profiling . Ranking & Serving 内容排序 – Personalized content based on user profiling Onstage Backstage User Profile Index Content Index Short-term User Behavioral Long-term User Behavioral Hadoop Pipeline (Batching) Ingest Normalize Feature Extraction User Modeling User Scoring Storm Pipeline (Streaming) Ingest Normalize Feature Extraction User Modeling User Scoring Content Processing Hadoop Pipeline (Batching) Feature Extraction Topic Modeling Content Event Pipeline Storm Pipeline (Streaming) Ingest Discover Feature Update Content Processing Storm Pipeline (Streaming) Ingest Normalize Feature Extraction HBase (Users) HBase (Content, Models) Ranking User Behavioral Signals Social Signals Content Agility Pipelining Build User Insight Build Personalized Content Gemini – 移动搜索和原生广告放在一起的服务 Yahoo! Proprietary and Confidential 18 The first unified ad marketplace for mobile search and native advertising. NEW UNIFIED CLOUD VISION 新统一云端愿景 Internet Serving Platform Supply-Side Platform Demand-Side Platform Backstage Onstage Ranking Indexing Emitting Pipelining User-based Profile Device-based Profile App-based Profile Personal Health & Fitness Pull Push & Pull & Stream Notification Serving Mobile Analytic Content Serving Ads Serving Search Query Serving Product Backend Advertising Marketplace Ad Exchange Data Management & Exchange Real-Time Bidding Ad Campaign Analytic Advertisers Users Web App Wearable App Developers Publishers Partners Twitter Facebook Yahoo! Proprietary and Confidential Mobile Services Internet Serving Platform Supply-Side Platform Demand-Side Platform Backstage Onstage Ranking Indexing Emitting Pipelining User-based Profile Device-based Profile App-based Profile Personal Health & Fitness Pull Push & Pull & Stream Notification Serving Mobile Analytic Content Serving Ads Serving Search Query Serving Product Backend Advertising Marketplace Ad Exchange Data Management & Exchange Real-Time Bidding Ad Campaign Analytic Advertisers Users Web App Wearable App Developers Publishers Partners Twitter Facebook Yahoo! Proprietary and Confidential Mobile Services User Social Signals Content Agility Internet Crawling Build User Insight Serve Personalized Content Personalization Pipeline Mobile Services OnePush Flurry Gemini User Understanding User Engagement Advertiser Engagement Developer Engagement Pipeline Processing User Behavioral Signals THANK YOU 谢谢大家 Notes & Acknowledgments 1. Internet of Things – Wikipedia, http://en.wikipedia.org/wiki/Internet_of_Things. 2. Liangjie Hong & Suju Rajan, “Science Powering Product and Personalization: Going Beyond Clicks”, Yahoo Lab, October 2014, http://yahoolabs.tumblr.com/post/99405569711/science-powering-product-and-personalization-going. 3. Michal Levin, “Designing Multi-Device Experiences – An Ecosystem Approach to User Experiences across Devices,” 2014, O’Reilly Yahoo! Proprietary and Confidential. 23
还剩22页未读

继续阅读

下载pdf到电脑,查找使用更方便

pdf的实际排版效果,会与网站的显示效果略有不同!!

需要 5 金币 [ 分享pdf获得金币 ] 0 人已下载

下载pdf