Download presentation
Presentation is loading. Please wait.
1
雷 凯 (leik@szpku.edu.cn) 北京大学深圳研究生院· 互联网研究中心 2012年5月8日 · 深圳 C 202
未来互联网体系构架专题 国内外未来互联网研究趋势 内容为中心网络(ICN)技术 几个(ICN)典型解决方案 重点关注CCN(NDN) 深圳实验室工作 国内其他高校工作进展 雷 凯 北京大学深圳研究生院· 互联网研究中心 2012年5月8日 · 深圳 C 202 北京大学深圳互联网中心
2
未来互联网体系构架(第一部分) 国内外未来互联网研究趋势
3
当前中国互联网发展趋势 日益增长的网络规模 截至2011年底,中国网站规模达到229.6万,较2010年底增长20%。
2011年底.CN域名注册量达到353万个较2011年中增长26000余个。 2012年1月16日,中国互联网络信息中心(CNNIC) 发布《第29次中国互联网络发展状况统计报告》
4
当前中国互联网发展趋势 网民人数规模已成平台化,仍然快速增加。
2012年1月16日,中国互联网络信息中心(CNNIC) 发布《第29次中国互联网络发展状况统计报告》
5
当前中国互联网应用发展趋势 2011年互联网应用习惯出现显著变化,包括新型即时通信、微 博等在内的新兴互联网应用迅速扩散,与此同时一些传统的网络 应用使用率明显下滑,显示出互联网新速度之快。 2011年中国网民即时通信使用率上升较快,增至80.9%。 电子邮件使用率从2010年的54.6%降至47.9%,用户量减 少392万人; 论坛/BBS则由32.4%降至28.2%, 用户量也略有减少。 社交网站使用率目前使用率止跌回稳, 达到47.6%。 娱乐类应用普遍走低,但网络视频用户增幅明显 网络音乐、网络游戏和网络文学用户规模在2011年增长幅度较小,使用率也 分别下滑至75.2%、63.2%和39.5%。相比之下网络视频行业的发展势头 相对良好,用户规模同比增加14.6%,达到3.25亿人,使用率提升至 63.4%。
6
当前中国互联网应用发展趋势 截至2011年12月底,我国微博用户数达到2.5亿,较上一年底增长了 296.0%,网民使用率为48.7%。微博用一年时间发展成为近一半中国 网民使用的重要互联网应用。网民使用率为48.7%。 截至2011年12月底,中国网络游戏用户规模达 到3.24亿,较去年同期的3.04亿增长6.6%。 网民使用比例为63.2%。 截至2011年12月底,我国博客和个人空间用户数量为3.19亿,较 2010年底增长2414万,增长率为8.2%。 网络支付:截至2011年12月底,我国使用网上支付的用户 规模达到1.67亿,使用率提升至32.5%。与2010年相比, 用户增长2957万,增长率为21.6%。 2011年,中国人民银行分别向三批服务企业发放 《支付业务许可证》。
7
当前互联网面临的问题 海量数据的爆炸式单增 中国网页数量为866亿个,比2010年同期增长44.3%。
8
当前互联网面临的问题 人均网络带宽的降低和人均带宽需求上升之间的矛盾
截至2011年12月底,我国IPv4地址数量为3.30亿,较2010年底增长19.0%。 国际出口带宽为1,389,529Mbps,较2011年6月增加17.5%,较2010年同期上涨26.4%。 中国静态和动态网页比例从2010年12月底的1.14:1上升至2.18:1,造成这一变化的原因包括:第一,许多网站为了对搜索引擎友好,将动态内容静态化;第二,大量动态页面存在暗网中,对于这些页面搜索引擎很难收录完全
9
当前互联网面临的问题 负载不均衡 网络交换和链接设备是宽带网络的技术瓶颈。 2011年我国各省市网民规模均有明显增长,
中国大陆31个省(市、自治区)中网民数量 超过千万规模的省份达到21个。同时,我国 互联网发展的地域差异依然延续, 北京市的互联网普及率已经超过七成,达到70.3%,而互联网普及程度较低的云南、江西、贵州等省份互联网普及率不到25%。 网络交换和链接设备是宽带网络的技术瓶颈。 网络瓶颈指的是影响网络传输性能及稳定性的一些相关因素,如网络拓扑结构、网络连接设备(交换机和集线器)、网线、网卡、服务器配置等 骨干网压力、网络出入口压力非常大 北京大学深圳互联网中心
10
当前互联网面临的问题 大量的重复(冗余)传输
WWW、微博、微视频等绝大部分内容分发的互联网应用都存在非常大量的重复传输。优酷2011年播放量过亿电视剧有30部。 P2P、CND, 云计算等技术也都存在大量的重复传输。 北京大学深圳互联网中心
11
主要特征
12
未来网络流量分布预计
13
未来互联网体系设计的指导思想 未来网络应该以内容作为网络设计和服务模式的核心.
并以内容的高效获取和带宽的节约(降低数据包重复传输率)为双重优化目标。 未来网络将以“低带宽、均负载”的需求提供“高智能、高效率”的内容共享信息服务为主要研究方向
14
国内外未来互联网研究的相关组织 下一代互联网(NGI.next generation Internet)
网络创新的全球化环境(GENI, Global environment for network innovation) 未来互联网设计(FIND, Future Internet Design) 欧盟第七研发框架计划(FP7, EC 7th framework program) 中国的“973”, “863”,国家科技支撑计划等。
15
(Information Centric Network)
未来互联网体系构架(第二部分) 内容为中心网络研究 (Information Centric Network) 典型技术
16
(1)位置与标识分离(locator/identifier split)
IP地址语义过载是当前互联网在命名和寻址体系结构设计中存在的主要缺陷. IP地址语义过载指的是在当前的TCP/IP协议中,IP地址既代表了网络节点的拓扑位置,又是节点的标识。 IP地址双重身份造成了网络在移动性、可扩展性、安全性等方面的问题。 关键问题: 完成Locator和Identifier两套名字空间之间的转换(映射服务、地址前缀聚合、地址绑定、和移动性支持(Mobile IP)) 模型: MIT的FARA(Reorganizing the Address Architecture) 。
17
(1)传统的位置与标识分离的解决方案 映射服务是位置与标识分离的核心 高效、可扩展的映射服务是关键
18
(1)内容为中心的位置与标识分离的方案 在以“内容”为中心的未来互联网中,由于“内容”取代“主机 ”成为网络的“头等实体”(first class entity),这一思想也被 扩展为内容位置与内容标识的分离。 例如,分层命名架构中增加了2个新的名字空间 – 标识数据/服务 的Service ID和识节点的 Entity ID,借此实现标识和位置的分 离。 DONA(Data Oriented Network Architecture )架构中用公共 密钥加数据标签的形式来命名数据,数据的标识与地址(位置)不 再——对应。 NetInf架构中分别用Information Object 和Data Object 来表 示抽象的信息和具体的二进制数据,这两类Object 都被赋予与地 址无关的、全局唯一的标识符。
19
(2)基于名字(不是位置)的路由 通常讨论的路由指的是IP层的路由。用户的内容请求经过“名字解析”转换为IP地址,然后路由器根据IP地址进行数据包的路由转发。这样的定位、路由机制对于以内容分发为主要业务的网络来说并不是最优的,所以在未来网络的研究中,出现了一类新的路由方式,称为基于名字的路由(Name based Routing)
20
(2)Stanford大学的TRIAD 斯坦福大学在2000年提出了一个支持名字路由的网络架构TRIAD。该网络在口层之上增设了一个“内容层”。内容层由支持命名的内容路由器(Content Routing)来执行基于名字的路由。这些内容路由器像常规的IP路由器和名字服务器那样运行,可以实现基于IP的路由和基于名字的路由。基于名字的路由需要在现有的Intemet上增加2个新的协议: 1)因特网名字转换协议(INPR, Internet Relay Protocol) cDl),根据中继节点维护的路由信息实现名字到地址的转换; 2)基于名字的路由协议(NBRP, Name-Based Routing Protocol), ,动态地更新中继节点上的路由信息,类似于边界网 关协议。
21
(2)其他基于名字(不是位置)的路由 在结构化P2P系统中,基于分布式哈希表(DHT, distributed hash table)的路由是近几年的研究热点 ,也出现了很多新的路由协议嘲,如Chord、Pastry 、CAN、KAD等。 在基于DHT的P2P系统中,哈希表的键值通常与文件 名关联,从这一点来看基于DHT的路由也是与内容直 接相关的。 在NDN、DONA、Netlnf这些以内容为中心的网络架 构中也都通过不同的方式实现了基于名字的路由。
22
(3)发布/订阅模式(Publisher/Subscribe)
发布/订阅系统是一种使分布式系统中各参与者能以“发布/订阅”的方式进行交互的中间件系统。 在发布/订阅系统中,信息的生产者和消费者之间所交互的信息被称为“事件”(event)。 信息的“生产者”将“事件”发送给“发布/订阅中间件”; 信息的“消费者”则向“发布/订阅中间件”发出一个“订阅条件”,表示对系统中的哪些事件感兴趣; “发布/订阅中间件”负责将所发布的事件及时、可靠地传送给所有对之感兴趣的信息消费者。信息的生产者称为发布者(publisher),信息的消费者称为订阅者(subscriber),发布者和订阅者统称为客户(client)。 发布订阅模式(Pub/Sub,publish/subscribe paradigm)近年来得到了广泛的关注例,其主要原因是该模式能使大范围的分布式交互有较低的耦合度。
23
(4)内容缓存(Content Caching)
为了达到缩短内容获取延时、均衡网络负载的目的,一种常用的 方法是使用缓存(caching)技术。缓存代表了这样一类技术:将先 前获取到的信息保存到某个存储设备中,为后面对相同信息的请 求提供便利p”。在很多场景中都可以看到缓存技术的应用,如浏 览器缓存、web代理缓存、应用程序缓存等。 根据目前对未来互联网研究的了解,但凡涉及减小网络流量,提 高响应速度这些问题,大家普遍的做法就是引入缓存机制。 在内容分发网络中,布置在网络边缘的缓存服务器会将以前从网 站服务器获取到的内容保存下来。当接受到用户的请求时,它们 先在缓存中查找符合条件的内容,如果找到,则将这些内容直接 返回给用户。缓存技术的使用极大地缩短了内容的响应时间。在 NDN、DONA、NetInf等项目中也使用了缓存机制,实现方式与 内容分发网络基本相同,只是用于缓存的设备、判断内容是否需 要缓存的具体条件存在差异。
24
(Information Centric Network)
未来互联网体系构架(第三部分) 内容为中心网络研究 (Information Centric Network) 若干典型解决方案,分为: 渐进式的(evolutionary) 革命式的(revolutionary)
25
对等网络P2P (渐进式1) 说到内容共享当然就不能不提P2P。P2P网络中每个节点既充当服务 器,为其他节点提供服务,同时也享用其他节点提供的服务。P2P具 有良好的可扩展性、健壮性,性价比高,能实现更好的网络负载均衡 当前P2P主要面临以下几方面问题:拓扑失配;网络病毒、不良信息 的传播;网络带宽;由节点的自私性导致的不合作。其中带宽是最为 严重的一个问题。带宽问题一方面是指P2P应用消耗了大量网络带宽 资源(50%以上);另一方面是指带宽的不对称:下载带宽远远高于上 传带宽。 尽管存在着上述问题,弹性好、扩展性强、服务效率高等优点使得 P2P仍有可能成为未来互联网中一种重要的内容分发方式 代表性: 天网Maze; 欧盟第七框架中的NAPA-WINE(以改进P2P流 媒体分发性能为目标)、 P2P_Next(以P2P技术为基础的互联网电视 广播系统); P4P (旨在加强服务供应商(ISP)与客户端程序的通信,降 低骨干网络传输压力和运营成本,并提高改良P2P文件传输的性能)
26
Uswarm (渐进式2) Unified Swarm(Uswarm)是一个基于目前P2P结构的下一代互 联网数据传播体系。Swam指的是P2P网络中那些下载某一资源文 件的主机节点而形成的簇,这些簇是以一种松耦合的方式进行连接 的。其中,以BitTorrrent为代表的P2P文件共享系统在一定程度 上解决了用户只下载不上传的自私行为所导致的系统不稳定现象, 如采用Tit-for-Tat的博弈机制可以使用户更多的上传资源,采用 将一个资源文件分成若干片,不同的片可能来自于不同的上传者, 这样可以保障每个用户的下载容量大于自己的上传容量。 过命名解析服务和元数据解析服务来定位用户所需要的资源地址。 名解析服务(IRS, Internet Resolution Service)通过统一的元 数据标引机制将互联网上以URL形式展标引的各种资源转换成元数 据标引的形式。元数据解析服务(MRS, meta-data resolution service)将这些元数据转换成真实的节点物理地址(地址无关)
27
内容分发网络(CDN) (渐进式3) 内容分发网(Content Delivery Network)的目的是在现有的Internet中增加一层新的网络架构,将网站的内容发布到最接近用户的网络“边缘”,使用户可以就近取得所需的内容,以解决Internet网络拥塞的状况,提高用户访问网站的响应速度。 CDN的关键技术包括内容路由、内容分发、内容存储和管理等。 当前,CDN处在各运营商的自行管理和建设下,彼此之间孤立,带来了高成本重复建设、存储资源浪费、内容标识不一致、内容管理机制不统一等多方面的问题。 另一方面,大型文件(以GB为计算单位的软件、游戏、媒体文件等)的访问下载、用户视频的上传分享这些新兴的内容分发需求使得边缘服务器再一次成为瓶颈,部署更多的服务器对于CDN来说已经无法经济有效地解决问题。 由于CDN存在上述问题,“下一代内容分发网络”的慨念应运而生。将P2P技术引入CDN,形成对等网辅助.引入P2P虽然提高了CDN的可扩展睦,但同时给稳定性和可靠l生带来了隐患,并使CDN中原本就繁琐的内容管理更加复杂。
28
DONA (革命性1) DONA (Data Oriented Network Architecture)美国Berkeley大学的一个研究项目。是用于取代现行DNS而被提出的一种全新的基于内容的网络体系结构。该体系的主要不同点是基于数据内容的命名、基于内容的命名解析和路由、内容的缓存和中间沙箱等。 为了保证数据的持久性、可达性(变换了IP地址仍可以找到)、可认证性(安全性)等性质,将数据命名为P:L的形式,其中P是public key的加密形式,用于安全认证,L是该数据的标签,每份数据都有1个唯一的标签。 为了解析和路由,DONA中设计了一种叫做解析处理器(RH,Resolution Handler)的中间节点,包含现存网络中的DNS服务器和路由器的功能。为了实现基于内容的路由,有2种命令查找和注册,新数据向RH发送注册命令,数据失效时向RH发送解除注册命令;用户向RH发送查找命令来获取数据。根据一定的规则决定是否进行缓存。这样的缓存机制可以减少路由的跳数和对链路的流量压力。 另外,RH可以作为中间沙箱,起到防火墙的作用,并对数据内容进行审查、过滤,可有效阻止有害信息的传播,提高网络的安全性。
29
NetInf (革命性2) Network of Information (NetInf) )是欧盟4WARD项目的1个子项目,它扩展了标识位置分离的思想,通过递归查找的方式实现信息对象和存储位置的解耦。 NetIInf的研究内容包括信息建模、名称解析、路由、存储、搜索等。 NetIInf独立于具体传输层协议,可以以覆盖网络的方式部署在现有的网络基础结构上,也可以与其他新的传输技术相结合(如 Generic Path),构成一个全新的网络架构。
30
NDN/CCN (革命性3) Named Data Network(NDN)是美国国家科学基金会 2010年9月宣布资助的一个项目。NDN的基本思想是 将信息(或称为内容、数据)本身视为网络中的“头等实 体”,是将当前基于主机(host-based)的网络结构转 变为基于内容(Content-based)的构架。 为了实现以内容为中心,NDN的解决方案是以内容的 “名字”代替IP地址,重新设计数据包的格式和各层协 议,试图替换现有的IP架构。 CCN (Content Centric Network)基本和NDN是一回 事,NDN是NSF的项目名称,CCN是PARC的项目名称 接下来,我们会重点介绍。
31
未来互联网体系构架(第四部分) 重点关注CCN(NDN)项目
32
7
33
ISP ISP 8
34
Communication Distribution Naming Memory Security Endpoints Stuff
Explicit; Storage and wires equivalent Memory Invisible, Limited Security Secure the process Secure the stuff 13
35
Communication Distribution Naming Memory Security Endpoints Stuff
Explicit; Storage and wires equivalent Memory Invisible, Limited Security Secure the process Secure the stuff 13
36
Communication Distribution Naming Memory Security Endpoints Stuff
Explicit; Storage and wires equivalent Memory Invisible, Limited Security Secure the process Secure the stuff 13
37
dst DST SRC src 14
38
FIB dst DST SRC src 15
39
X Path determined by global routing, not local choice.
FIB dst DST SRC src Path determined by global routing, not local choice. Structural asymmetry precludes market mechanisms and encourages monopoly formation. 16
40
our approach 17
41
FIB Producer Content Store c /b / ? a Consumer 18
42
FIB a/b/c/d Producer Content Store a/b/c/d c /b / ? a Data Consumer 19
43
FIB a/b Producer Content Store /e b/c a/ ? Consumer 20
44
• Forwarding Upstream Packets say ‘what’ decision not is ‘who’ local
FIB a/b Producer Content Store /e b/c a/ ? Consumer • Forwarding decision is local Upstream performance measurable Packets say ‘what’ not ‘who’ (no src or dst) 21
45
ISP ISP 22
46
ISP ISP 23
47
Networking Named Data
48
Networking Named Data
49
Networking Named Data 三个主要的数据结构 the FIB (Forwarding Information Base)
Content Store (buffer memory) PIT (Pending Interest Table).
50
Networking Named Data
51
Networking Named Data Two Critical Difference of CCN: Strategy
Security
52
Networking Named Data Two Critical Difference of CCN: Strategy
The strategy layer (Section 3.3) makes the fine-grained, dynamic optimization choices needed to best exploit multiple connectivities under changing conditions. Security
53
Networking Named Data Strategy
Machines today typically have multiple network interfaces and are increasingly mobile. Since IP is restricted to forwarding on spanning trees, it is difficult for IP to take advantage of more than one interface or adapt to the changes produced by rapid mobility. CCN packets cannot loop so CCN can take full advantage of multiple interfaces. CCN talks about data, not to nodes, so it does not need to obtain or bind a layer 3 identity (IP address) to a layer 2 identity such as a MAC address. Even when connectivity is rapidly changing, CCN can always exchange data as soon as it is physically possible to do so. Furthermore, since CCN Interests and Data are paired, each node gets fine grained, per-prefix, per-face performance information for adaptively choosing the ‘best’ face for forwarding Interests matching some prefix
54
Securing (digital) stuff
Packet = 〈 name, data, signature 〉 Any consumer can assess solely from the data: •完整性 Integrity (is data intact and complete?) 相关性 Pertinence (is this an answer to my question?) 来源可信度Provenance (who asserts this is an answer?) 24
55
Evidentiary Trust A rich web of trustworthy information
arises from named, signed data: Name Hierarchy & Links Content ↕ Key Content ↕ Key Key Certification Graph 25
56
Evidentiary Trust A rich web of trustworthy information
arises from named, signed data: Name Hierarchy & Links Content ↕ Key Content ↕ Key Key Certification Graph Content ↕ Key Content ↕ Key Content ↕ Key Content ↕ Key 25
57
Evidentiary Trust • Attacker’s job gets exponentially harder as you
accumulate information. ➡ Security is emergent (自然发生) property of the system. Name Hierarchy & Links Content ↕ Key Content ↕ Key Key Certification Graph Content ↕ Key Content ↕ Key Content ↕ Key Content ↕ Key 26
58
Research Agenda 27
59
Application Middleware NDN TCP/IP Information-to-Conversation
model translation TCP/IP NDN 28
60
Application Application Middleware NDN TCP/IP
Information-to-Conversation model translation TCP/IP NDN 28
61
NDN Core 29
62
Content Participatory Distribution Sensing NDN Core Fast Forwarding
Instrumented Environments Content Distribution Participatory Sensing NDN Core Fast Forwarding 29
63
• Any-to-Any communication • Bandwidth / Memory / Distance tradeoffs
Instrumented Environments Content Distribution Participatory Sensing NDN Core Fast Forwarding Security • Fast signing • Usable Trust • Privacy • Attack resistance Routing • Existing models • New models Fundamental Theory • Any-to-Any communication • Bandwidth / Memory / Distance tradeoffs 29
64
Kai Lei (雷 凯) leik@szpku.edu.cn May 2nd, 2012
The 3rd Annual JRI Symposium (UCLA and PKU) Research and practice of CCN on P2P and video streaming -our understanding, approaches and targets Good Afternoon, everyone. Dear Faculty, Friends and students, I am happy here to attend this JRI meeting, and my name is Kai Lei, I come from Center for Internet Research and Engineering, PKUSZ. And the talk that I am going to give today is “…” . Kai Lei (雷 凯) May 2nd, 2012
65
Presentation Outlines
Our Participation history of CCN community Our understanding about CCN, Why are we combining P2P and CCN together? Our approaches : develop open internet application as a research platform, to train students and gather experimental data Our targets: Develop one P2P like file sharing system (CCNMaze) over CCN Develop another P2P like video streaming system over CCN (live and Vod) Our exploration: It seemed that we have many kinds of problems that we can study with CCN. But what are the most important ones, good for us.
66
Brief about CCN Content Centric Network, belongs to Future Internet Architecture research area. It was also called as Named Data Network project, Led by Prof. Lixia Zhang and other schools in US. (NSF project started from Sep. 2010) Not location based like TCP/IP Interest Driven Decentralized Data level Security, not System level Data has “life cycle”
67
Get in Touch with CCN (4 stages)
Got to Know about CCN and Study CCN (Oct,2010) Reading Papers, build and install CCN software Prepare research resources for CCN (Jan, 2011) 国家自然基金《基于CCN的新互联网架构体系对比分析及其路由缓冲策略研究》, 项目编号:NSFC , 2011年1月-2014年12月 Collecting papers, Student Encouragement Extend international connections with CCN community (July, 2011, Sep, 2011) CCNxCon2011, Prof. Lixia Zhang, Van Jackson and other PARC People, Finished some primitive work and papers (now) CCNMaze, CCNHippo CCNMaze paper (Rejected by InfoComm NoMen Workshop) CA security for CCNMaze Paper
68
Why are we interested in CCN? (1)
Comparison P2P CCN Network Protocol Connection Based (TCP/IP) Content based (Interest Driven) Architecture Half Decentralized (Still need Central Svrs) Complete Serverless Data Distribution* (details in next 2nd) P2P at Application Level Transportation Level (more general purpose) Security Passive data protection Additional system level protection needed, hard to identify . Named Data (easy to Identify) Inherited Data protection itself
69
Why are we interested in CCN? (2)
Comparison P2P CCN Bandwidth usage Massive data duplication , Worse congestion Nearest Data first, Smart data replication Copyright Control No feasible ways Resistant by carriers and operators Data Marked by name, good for data control and management Business Model Outdated (say Napster, BT and Xunlei,etc) New routers, New OS, New Applications, etc. Green Computing P2P worsens the problem, Greedy consumption Excessive usage Great saving of resources, especially for multimedia data and sharing applications
70
*CCN benefits P2P at transportation level
CCN constructs a P2P like content distribution at router level, can greatly improve content distribution, reduce traffic data redundancy. CCN shortens the data transfer path between peers
71
IP based P2P Systems in CIRE, PKU
Maze system ( since 2005) Once the most popular P2P system file sharing system in Chinese education network (CERNET). A small social network platform A great excising and experiment platform for student training. Hippo live streaming system (since 2009) Serve for 3 school campuses’ (PKU, THU and HIT) faculty and students (6k people) at University Town in Shenzhen. High Definition Video Streaming (live and VOD) Broadcasted 2010 South Africa World Soccer Games. Plan to Broadcast 2012 London Olympic Games this summer.
72
Why Maze works better with CCN ?
Decentralization can reduce Maze Central Servers’ load Connection- Independency can improve Peers’ connectivity (NAT) DHT like Interests Driven communication mode can improve searching Data-level security is good for anti-virus, File Spamming and Privacy
73
What can Maze benefit CCN
So far, (according to our knowledge) there is no real CCN running system for large volume file sharing and transferring Number of Maze users can gather real internet data to CCN benchmarking, test and estimate it scalability, etc. Our previous research of P2P could promote or contribute to CCN (from application point of view)
74
Working Progress CCNMaze
Implemented based on software CCND, over CCN protocols (IP overlay) Functions: User Registration and Authentication , User Discovery, file sharing and downloading, file searching, Friends Management, etc. Deployment: currently only within our Lab, aim to go public Running Status: Reduced number of central servers, file transferring is very good.
75
Save us a few servers - easier to manage
New User Server User Server HB Server State Server Index Server
76
To do Next Complete CCNMaze’s functionalities for public downloading and usage. Aim to Integrate with current IP Maze, if possible. Try to introduce Maze to Mobile Network Collect running statics (Log analysis)
77
System pictures (a simple Java GUI)
GUI Add Sharing Directories GUI - 2. Browsing a Directory
78
System pictures GUI- 3. file Searching GUI – 4. Searching Results
79
System pictures GUI - 5. Download a file GUI - 6. Downloading status
80
A major problem encounted
While deploying CCN Internet App (Routing) Solution in small scale network (Lab) We use some PCs to simulate a CCN network CCND is statically pre-configured (Manually set the routing path and network topology) Ideas in the future When real CCN routers come into life, these worries would be gone.
81
Routing Solution for CCNMaze
Targets: for real normal internet users to use with existing TCP/IP Internet. Comparison with current IP version of maze. A realistic difficulty No feasible to arrange enough CCND nodes. Tentative solution : Place some servers as fixed CCND nodes (like Skype’s Super nodes) Selecting some user clients’ PCs to run CCND
82
Routing solution for CCNMaze (Conti.)
Problems may be faced For CCND of User node, how to plan its next CCND node? How to deal with User nodes’ on and off ? User nodes selection need favor of nearest ones. We had researched on IP Maze about “Network Neighbors” , “self –organized User Zone” and DHT. We will add a routing servers for CCNMaze. User log on an off. Client detects its neighbor and verifies them periodically. Routing server provides alternatives for client to re-plan Node Ranking algorithm. (based history info and system status)
83
Research problem A: User Distance
Algorithm or methods to calculate User Distance (by speed or by hops or by throughput?) . Not only its network attributes, but also computing power and disk space, etc. This a typical P2P research problems, but with some additional requirements for CCN.
84
Problem B: Data Storage and Cache
CCN maps a file in OS into segments in CCN Repository. The format of the same data was changed. On the disk:OS Data blocks (NTFS or Linux) CCN Repo:Named Data segments Possible data redundancy (duplication cause more disk usage) Size limitation of Repo (Cloud solution?) The size of Repo must have a upper limit in real system.
85
The Repo of CCNMaze Move sharing files into Repo for fast transferring. Only those files to be transferred will be written into Repo. Those files got transferred more will benefit more from Repo. Side Impact: files exist multiple copies (disks, repo, multi nodes ), increased the overall storage requests. It leads to users may reluctant to share or reduce number of sharing files. Compared to VOD, the reward from file sharing is smaller How to identify “Hot files” is necessary to guide repo replacement strategies for the whole system. Let’s look at how we deal with Repo in CCNMaze
86
Video Streaming over CCN
Live Streaming and Video on Demand (VOD) application over CCN -- Use our own Hippo System OK, here comes to anther topic, as I mentioned before, we also want to do some video streaming work over CCN.
87
Hippo - Video Streaming over TCP/IP
P2P + CDN Segmented Content isolated with service Zone Multi tier
88
CCNHippo Compared to file sharing, video streaming is more real time, which can fit better into CCN concepts. The implementation experience of CCNMaze could help us speed up the development of CCNHippo Targets: First implement the basic functionalities of video streaming system. Inter-communication between CCNMaze and CCNHippo. Comparison between CCNMaze and CCNHippo to find some interesting conclustions.
89
Initiatives Why implement Hippo over CCN
Video Streaming system requires more real-time data distribution, the mode of CCN can fully take advantages of content centric philosophy and greatly save the network bandwidth. CCN named data approaches could provide solutions to copyright and accounting issues of multimedia applications.
90
Initial Design of CCNHippo
Use CCN’s Repo as resource pool, providing sharing service, both live or VOD contents will be written into repository, including Static Node (CDN servers, so called Speer ) and Dpeer (Dynamic peer – user node) User Gstreamer plug-in to request、read and play video blocks in CCN network, local repo provides sharing functionality for others. Macro management of the data in all repos (both Speer and Dpeer), in order to achieve overall system QoS (just like SM server in IP version of HIPPO. ContentStore is still managed by CCN itself.
91
Research Problem C: Hippo with CCN
Similarity between Hippo and CCN Hippo used segments for video contents, similar to CCN The QoS of Hippo relies on caching replacement strategies, similar to CCN ‘s contentstore. Difference between Hippo and CCN Hippo has a RM (Regional Server) to manage partner nodes relationship, the routing of CCN is not certain yet, not easy to identify nodes nearby. Hippo considers caches in both memory and disks, while ContentStore in memory and repo in disks, need determine caching strategies. Focus: Build application-specific routing strategies, self-organize nodes structure and network topology. Design a application-specific caching replacement strategies, For live application, consider to design video streaming specific cache in memory in contentstore. For VOD, consider combining ContentStore and Repository together , design a multi-tier data caching mechanism
92
Security : CA based authentication for CCNMaze Users
Another small piece of work Security : CA based authentication for CCNMaze Users
93
Comparison of Existing Trust Models
Model Name Advantages Disadvantages PKI/CA Unified standards Strong authentication Widely applied Hard to globally deploy due to political reasons Uncertain transitive trust WOT Decentralization Leverage social connections Hard to be subverted Lack of introducers Isolated peers, cold start Weak ties between peers Key management problems SPKI/SDSI Regional autonomy Scalability Scarce links between groups Various criteria of trust (hard to be standardized)
94
CA-based Trust Architecture
Multi-level CA fit for the hierarchical structure of CCN name Currently 2-level (at most 3) Each organization has its own CA server, which is responsible for user registration, certificate issuing and signing. The registered user and certificates should be reported to the higher-level CA. Two core functionalities of CA: Issuing certificates Maintaining credit records of users The basic functionality of Internet security is to associate digital contents with real entities by issuing certificates. We’d rather not to judge a user is trustworthy or not, the trustworthy of a user should be judged contextually by application or users themselvs. * /pku.edu /pku.edu/CIRE / CA of PKU.edu CA of CIRE Root CA
95
Thinking Leverage the advantages of CCN more in CA for the CCNmaze system? We tried to make the distribution procedure of certificates faster and increased the copies of certificates in the network. -> by using CCN content store and cache mechanism Version control of certificates (by dates) -> by using CCN content version control Authenticate the user can avoid spamming.
96
未来互联网体系构架(第六部分) 国内其他高校工作进展
97
973项目“面向服务的未来互联网体系结构与机制研究”
现行互联网是基于TCP/IP体系结构建立的,其假设用户和终端是 可信和智能的,网络本身仅仅需要提供尽力而为的数据包转发服 务,这种理念符合最初以主机互联和资源共享为主要目标的互联 网设计需求。随着应用及计算模式的日益丰富及社会对互联网依 赖程度的增强,互联网接入方式和网络功能定位发生了巨大的改 变,TCP/IP体系结构已经无法满足互联网持续发展的需求,在可 扩展性、动态性以及安全可控性等方面呈现出无法解决的问题。 为了从根本上解决上述问题,适应互联网未来持续发展的需求, 需要设计新的互联网体系结构并研究相关关键机理。 中国科学院计算技术研究所、北京邮电大学、清华大学、中国科 学院计算机网络信息中心、重庆邮电大学、西安电子科技大学、 中国科学院软件研究所、数据通信科学技术研究所、湖南大学、 中国联合网络通信有限公司、中国人民解放军理工大学。
98
清华大学 – 毕军教授 A NDN testbed with real traffic via Http‐NDN gateway
For research experiment: to build a large scale testbed with NDN traffic that is easily generated by a Http‐NDN gateway (as http proxy for students) from translating students’ daily http traffic (short‐term) For NDN dissemination: To make full use of NDN’s benefits (such as caching in NDN router) to improve student’s web browsing performance, and/or to save campus operators bandwidth (mid‐term) User/server transparency is good for experiments No change for HTTP browser/server Easily deployment The match between HTTP “semantics” and CCN is pretty decent To simply do “syntax” translation can achieve most “semantics” Translation is easy for simple browsing task Gateway still need to fully understand HTTP “semantics” Not just delivering packets by “syntax” translation, for advanced browsing task
99
CCN相关的人员介绍 关于京大学-洛杉矶加州大学理工联合研究所
北京大学-洛杉矶加州大学理工联合研究所(以下简称JRI)成立于2009年6月16日,由北京大学的周其凤校长和UCLA的Gene Block校长共同签署合作协议。JRI旨在促进两校在理工及医学领域的学术研究和人才培养的提高。 JRI由北京大学的李晓明教授和UCLA的丛京生教授共同担任联合主任。截至目前,已有来自各理工院系及医学院的59位北大教授和74位UCLA教授加入JRI。
100
CCN相关的人员介绍 张丽霞教授个人网站: http://www.cs.ucla.edu/~lixia/
张北川教授个人网站: 王岚教授个人网站:
101
CCN相关的人员介绍 刘韵洁院士简介: 山东烟台人。1968年毕业于北京大学技术物理系。曾任邮电部数据所所长、邮电部电信总局副局长兼数据通信局局长、中国联通总工程师、副总裁等职,现任中国联通科技委主任。 曾主持规划、设计、建设和运营管理了中国公用数据网,其中包括公用分组交换网(CHINAPAC)、公用数字数据网(CHINADDN)、公用宽带网(CHINA FR&ATM)、公用互联网(CHINANET)。开拓了我国公用数据通信新领域,形成我国上千亿的网络和应用产业,为我国信息化发展作出突出贡献, 1998年曾被美国时代周刊评为全球IT领域最重要的50位杰出人物之一(列28位)。 先后获得国家科技进步一等奖1项,部级科技进步一等奖2项,国家发明专利多项,先后发表学术论文59篇,专著7部。
102
相关CCN的资料 CCN 官方网站: http://www.ccnx.org
NDN project 官方网站: CCN 工作组: CCN Video Streaming 相关研究(Future Multimedia Networking Systems) InfoCom 2012, 美国佛罗里达,第一届NDN workshop ( SIGCOMM 2012, 第一届 ICN WorkShop ( Paper submission :Sunday, March 25, 2012 [23:59 PST] 2011 ACM SIGCOMM ICN,TORONTO, CANADA (
Similar presentations