On Maximizing Tree Bandwidth for Topology-Aware Peer-to-Peer Streaming
一种依赖社会网络的P2P视频点播推荐机制和振华

小型微型计算机系统Journal of Chinese Computer Systems 2013年2月第2期Vol.34No.22013收稿日期:2011-07-21收修改稿日期:2011-10-27基金项目:安徽省自然科学基金项目(11040606Q52)资助;中央高校基本科研业务专项(WK0110000024)资助.作者简介:和振华,男,1986年生,硕士研究生,研究方向为P2P 系统;田野,男,1977年生,副教授,研究方向为复杂系统分析、P2P 系统、网络测量等.一种依赖社会网络的P2P 视频点播推荐机制和振华,田野(安徽省高性能计算重点实验室,合肥230027)(中国科学技术大学计算机科学与技术学院,合肥230026)E-mail :ehzh@mail.ustc.edu.cn ;yetian@vstc.edo.cn摘要:提出一种新颖实用的视频推荐机制,用于缓解高质量P2P 视频点播服务中存在的节点带宽饥饿现象.通过分析中国科技大学校园视频点播系统10个月的点播日志,发现视频在视频关联网络有聚簇的趋势,社会网络的现象明显.基于视频的社会网络特性,利用P2P 视频点播系统已有的gossip 协议,设计了一种新颖实用的基于社会网络的分布式视频推荐机制.利用gossip 协议和视频之间的社会网络关系属性,此机制能够分布式的并且低代价的给饥饿节点推荐最优的视频.饥饿的节点通过接受推荐来避免继续陷入饥饿.实验表明该推荐机制能有效地缓解点播节点的带宽饥饿,提高P2P 网络资源的利用效率.关键词:视频点播;分布式;推荐机制;社会网络中图分类号:TP393文献标识码:A文章编号:1000-1220(2013)02-0229-04Novel Video Recommendation Mechanism for P2P VoD ServicesHE Zhen-hua ,TIAN Ye(Anhui Key Laboratory on High Performance Compoting ,Hefei 230027,China )(School of Computer Science and Technology ,University of Science and Technology of China ,Hefei 230026,China )Abstract :A practical and novel social network-based video recommendation mechanism is proposed in this paper to alleviate the starva-tion in peer-to-peer (P2P )video-on-demand (VoD )services caused by imbalanced peer bandwidths and high-quality video requests.We analyzed log files collected from a campus-wide video-on-demand service of University of Science and Technology of China ,and ob-serve that videos requested by users'tend to cluster together in the network of video association and exhibit strong social network features.We propose a distributed video recommendation mechanism that piggybacks on the gossip protocol of P2P VoD systems.By piggybacking on the gossip protocol and exploiting the social networking property among the videos ,the mechanism can recommend videos to starving peers in a distributed and inexpensive way.And by switching to a recommended video ,a starving peer can avoid channels that are under starvation.At the same time ,the mechanism can effectively enable more successful video requests and alleviate the starvation problem by recommending videos that are not under starvation to interested starving users.Simulation-based experiments show that the mechanism can alleviate the starvation effectively and inexpensively ,and improve P2P network's resource utilization.Key words :Video-on-Demand ;distributed system ;recommendation system ;social network1引言基于P2P 网络的视频点播服务是目前互联网上视频点播的主要方式,典型的系统有PPTV 、PPStream 和UUSee 等.此外,随着多媒体技术的发展,高清编码的视频内容已相当普及.然而,在现有的P2P 视频点播系统中,点播高清编码的视频内容还存在许多困难[1].这些困难主要有:1)需要更高的上传带宽支持.ADSL 是目前我国互联网接入的普遍方式,而对于ADSL 用户来说,其最大上传带宽一般会小于400kbps[2].而常见的高清数字视频流媒体通常需要兆比特级别的带宽.当用户不能从P2P 系统获得足够的带宽下载所请求的视频时,会产生带宽饥饿现象[3].文献[3]通过测量p2ptv ,发现整个P2P 系统的带宽需求不平衡.从而得知需要更合理地分配节点的上传带宽资源来防止饥饿的产生[4].2)更大的缓存空间要求.高清视频不仅需要良好的上载带宽[4],而且需要用户节点提供更大的存储容量.相对于高清编码的视频,用户所提供的缓存空间很小,显然对P2P 系统的贡献就会很少.需要更合理地分配节点上的缓存资源[5].文献[6]指出P2P 视频点播服务的最大容量受限于被过分点播的视频内容,当点播需求超过最大容量,则发生带宽饥饿;优化节点的带宽分配可以提高系统容量.另一方面,P2P 视频点播系统中的用户节点具有自私特性,系统难以直接调度节点的视频缓存和上传来优化其带宽分配[7].近年来,一些研究发现P2P系统中的用户行为具有很强的社会网络特性[8],利用用户的社会网络性,P2P系统的性能可以得到进一步优化.在我们的工作中,通过对中国科技大学校园视频点播系统[9]10个月点播日志实施数据挖掘,我们分析了用户在视频点播中的行为特性,并构建了一个视频关联图模型.借助视频关联图,我们发现在用户在视频请求中存在明显的社会网络特性[3].基于这一发现,本文设计了一个应用于P2P高质量视频点播服务的、基于社会网络的视频推荐机制.在视频推荐机制中:当某一节点在请求某一视频时,如果由于缓存和上传带宽资源的限制,用户不能获得足够下载带宽而产生饥饿现象,则用户通过分布式的推荐机制,获取当前资源充足且感兴趣的视频列表;用户通过转而请求推荐的视频内容以避免饥饿.仿真实验表明该推荐机制可以有效地缓解P2P视频点播中节点的饥饿现象,提高P2P视频点播系统中资源的利用率.2视频点播用户行为挖掘与分析我们对中国科技大学校园视频点播系统(USTC VoD)从2009年4月29日到2010年2月27日大约10个月的点播日志[11]实施数据挖掘与分析.USTC VoD采用Microsoft的Win-dows Media Services(WMS)作为点播服务平台.在WMS中,当用户请求一个视频,或进行一次VCR操作,比如快进、暂停或开始,系统都会产生日志信息.USTC VoD系统包含超过15,000个视频供校园网用户点播.USTC VoD系统包括2台点播服务器,包含不同视频内容:1)教育类视频,2)影视娱乐类视频.为了避免视频内容的干扰而更好的观察用户的行为,所以我们仅考虑用户对单个视频内容的点播行为,而不考虑有承接关系的视频,如电视连续剧等.我们共观察到5,819个用户对2,651个教育类视频的53,516次点播和5,957个用户对3,082个娱乐类视频的70,772次点播.2.1视频关联图模型和社会网络性对用户请求视频的行为进行建模,我们可以获得视频之间的关联程度.现引入有向权重图G=(V,E)命名为视频关联图模型.数据集中的每个视频对应关联图模型G上的一个节点.针对任意两个视频i,j,其对应到图模型上的节点定义为v i,v j.边e ij上的权重定义为ωij,ωij=既请求视频i又请求视频j的用户对i,j的请求次数请求视频i的用户对i的请求次数(1)对视频关联图模型G,我们有如下观察:观察1:对于USTC VoD数据集中的某一个用户,其请求视频的序列可以视为在关联图模型G上的一条游走路径.观察2:如果打乱某个用户的已知视频点播顺序,那么我们仍然可以在关联图模型G上找到一条游走路径,并且路径的节点顺序与对应的视频点播序列相同.基于以上观察,我们发现用户请求视频的序列可以看作是关联图模型G上的一条游走路径.假设视频之间没有承接关系(例如不包含电视剧集等视频内容),则可以把用户的点播行为描述为视频关联图模型G上的随机游走.我们关注的是关联图模型G中视频对应节点之间的关系.关联系数(Clustering coefficient)是在图论中用于描述图中节点之间聚簇程度的关键指标.由于本文定义的关联图模型G为一个有向权重图,因此采用Fagiolo[12]的定义来计算关联系数.关联图模型G上节点和图的关联系数定义如下:1)对节点vi,则它与它的邻居节点间能够形成有向三角形数目为:T Di=∑(aij+aji)·(∑j≠i(aij+aji)-1)-2∑j≠iaijaji(2)其中{a ij}为关联图模型G的矩阵描述.2)节点vi与它的邻居节点间实际形成的有向三角形的个数为:(^W+^W T)3ii=({w1/3ij}+{w1/3ji})3ii=∑i≠j∑h≠(i,j)[(w1/3ij+w1/3ji)(w1/3ih+w1/3hi)(w1/3jh+w1/3hj)](3)3)节点vi的关联系数为^C Di=(^W+^W T)3ii2T Di(4)4)包含N个节点的有向权重图的关联系数为C=∑^C D iN(5)利用USTC VoD数据集中用户对教育类视频和娱乐类视频这两类视频内容的点播行为构造视频关联图模型,并分析图的关联系数.对于每一类视频,我们利用USTC VoD数据集中用户对教育类视频和娱乐类视频这两类视频内容的点播行为构造视频关联图模型,并分析图的关联系数,如图1.对于每一类视频,我们分别构造了3个视频关联图模型,分别包含用户在1个月、2个月、3个月内的视频点播请求行为.图1视频关联图模型的关联系数Fig.1Clustering coefficient另外,针对每个关联图模型图,在节点个数、节点的平均度和有向边的平均权重系数不变的条件下,我们分别构造了有向权重随机图,并计算图的关联系数.图1显示了一、二和三个月以来,根据用户对两类视频的点播行为所构造的视频关联图模型,以及对应的有向权重随机图模型的关联系数.从图1中可以看出,视频关联图模型的关联系数远大于随机图模型,表明节点对应的视频随用户点播请求具有很强的关联趋势.此外,我们还发现:1)娱乐类视频的关联图关联系数大于教育类视频.这很可能是由于用户在娱乐上的兴趣影响比在科教方面要深刻.2)数据集的时间跨度越大,其关联系数值反而越小,也即节点开始趋于分散.032小型微型计算机系统2013年3P2P视频点播视频推荐机制设计由于点播节点在上载带宽以及存储上的不均衡性,导致现有的P2P系统对高清视频支持能力有限,当点播节点不能够从P2P系统中获得足够带宽时,即产生带宽饥饿.基于此,我们提出了一种分布式的基于社会网络的视频推荐机制,旨在最大化地利用P2P系统的带宽和存储资源,避免点播节点饥饿.我们的主要思路是,当用户点播某一视频资源时,如果产生带宽饥饿,则系统根据用户此时的点播请求内容推荐给用户若干视频资源.而推选的资源应该满足以下两个条件:1)符合用户当前的兴趣;2)此资源在P2P系统中有丰富的缓存和下载带宽.3.1推荐机制设计我们的分布式推荐机制由节点上的推荐模块组成,介绍如下:1)处理视频请求历史在包含推荐机制的P2P点播系统中,每个节点都会保存自己的视频请求信息.一个节点的请求历史假设保存在一个长度为L的FIFO队列中.用户每成功点播一个视频,相应的视频条目就会追加到队列中.对某个节点x,它的历史队列记为h x.我们利用P2P流媒体系统中广泛使用的gossip协议[13]来交换和共享节点间的视频请求历史信息,并获取推荐信息.在gossip协议中,节点周期性地与它的邻居节点交换gossip 报文,包含如节点上本地缓存的视频块信息、节点的邻居信息等.通过gossip协议,节点可以有效地定位所请求视频的资源节点.对于节点x,定义它的邻居节点集合为N x.当节点x在观看视频i时,其通过gossip发现它的邻居节点y,此时,两节点交换它们的历史队列h x和h y.通过与所有邻居节点交换gossip报文,节点x获得了其邻居节点的点播请求历史记录集合,记为H x={h y}y∈Nx.当节点请求视频i 时,如果不能从P2P视频点播系统获得足够的带宽下载视频内容,则x发生饥饿,此时节点上的推荐模块启动,通过查询邻居节点点播历史集合H x来定位一系列用户可能感兴趣的视频,并且确保获得推荐的视频在当前P2P点播系统中有充足的资源可供节点x下载.推荐模块工作流程如下:a)获取候选视频集合:首先初始化候选视频集合C,以视频i为关键字遍历其邻居节点的点播请求历史记录集合H x={hy }y∈N x,若视频i不等于视频j,并且视频i与视频j均在点播请求历史记录集合H x中,则将视频j添加到候选视频集合C中.至此,完成候选视频集合获取;b)对候选集合评分:遍历整个候选视频集合C,并对集合内的每个视频利用2)部分的评分策略进行评分;c)形成推荐列表R:根据评分值对候选视频集合C中的元素进行排序,并选出k前项,然后推送给处于饥饿中的节点.2)评分策略在上述视频推荐过程中,对每个候选视频,我们计算一个分值,用于表示节点x对候选视频感兴趣的程度.我们考虑3种计算方法:基于距离的评分策略、基于相似度的评分策略、基于距离和相似度的混合评分策略.a)基于点播请求距离S(c,i)=∑h y:i∈h ye-dis(h y:i,c)(6)距离dis(h y:i,c)在这里表示在包含自对视频i的请求的历史记录h y中,从请求视频i完到请求视频c之间的视频点播请求条目个数.b)基于节点相似度S(c,i)=∑h y:i∈h ysim(x,y)(7)在这里,sim(x,y)表示节点x与节点y请求历史的相似度.对于相似度,可以通过计算节点x与节点y请求历史记录的皮尔森相关系数(Pearson correlation)或者余弦相似度获得.c)混合评分策略S(c,i)=∑h y:i∈h ysim(x,y)ˑe-dis(h y:i,c)(8)这里对每一个候选视频c,我们综合节点相似度和视频点播请求距离计算候选视频的分值.4仿真实验与性能分析为了检验推荐机制的有效性,本文进行了仿真实验.在实验中,我们使用教育类数据集来模拟用户请求行为.首先,根据视频的流行度随机的给用户初始化一个请求视频,然后在关联图模型G上以节点关联系数的概率随机游走,以此模拟用户的点播行为.也即,当用户观看完一个视频后,他会随机地以当前节点关联系数的概率去请求下一个视频.假设所有的视频都是同样的编码率,当用户请求某一视频时,所有的节点和服务器均尽最大努力提供上载带宽.然而当用户请求不到足够的带宽时,此时,节点被认为是处于饥饿状态.此时如有推荐机制,用户就可以以一定的概率请求推荐的视频.所以,我们考虑以下实验方法:1)没推荐机制:整个P2P系统中不提供任何推荐机制.2)随机推荐机制:在随机推荐策略中,当用户请求失败后,随机从系统中选取部分视频作为推荐视频列表.3)基于社会网络的分布式推荐机制:在仿真系统中,我们选用基于距离和相似度的混合评分策略作为推荐机制的评分策略,以此为依据产生推荐视频列表.4)理想系统:在这个理想策略中,我们假设整个P2P视频点播系统拥有关联图模型G的全局信息.并且当用户请求失败后,理想系统优先推荐给用户评分最高的视频.显然,这只是理想系统,在分布式环境中没有太大的实用价值,但是我们可以把它当作参照对象,来检验基于社会网络推荐机制的有效性.在仿真系统中,当用户获得推荐视频列表后,我们通过全局的视频关联图模型判断用户是否对推荐的视频感兴趣.当用户不能从推荐列表中获得感兴趣的视频,则该用户的点播请求失败;而无论用户成功点播其初始选择的视频或者任意推荐视频,我们认为该用户的点播请求成功.我们通过计算不同P2P视频点播系统下点播请求成功的次数来衡量推荐机制的性能.我们在仿真系统中模拟10,000个用户节点,并对用户节点的下载带宽不做限制.视频服务器的上载带宽被设定为1322期和振华等:一种依赖社会网络的P2P视频点播推荐机制(视频数量+1000)ˑ视频编码率.图2显示了不同策略的对比情况.从图2可以看到在随机推荐机制下,系统中用户节点的成功请求次数仅略高于无推荐的P2P点播系统.然而,对比随图2不同策略下请求成功数量Fig.2Number of successful requests with different policy机推荐机制和基于社会网络的分布式推荐机制,可以看出基于社会网络的推荐机制通过分布式的视频推荐机制,有效地利用了P2P网络的带宽和存储资源,提高了视频点播请求的成功率.表1推荐的质量和数量Table1Quality and quantity of recommendations平均边权重接受推荐次数随机推荐机制0.0031847,168理想系统0.014718105,048基于社会网络的推荐机制0.00701357,710同时,我们还关注推荐机制的准确性和推荐视频的数量.表1列出了在三种推荐机制下,用户接受推荐视频的次数,以及所接受的推荐视频与用户初始选择视频在视频关联图模型G上对应的有向边的平均权重.从表1可以看出:1)理想系统推荐的视频之间具有较高的平均权重,这表明推荐的视频有较大可能被用户接受;2)与随机推荐相比,基于社会网络的推荐机制更能推荐平均权重较大的视频,因此,获得用户接受的概率比随机策略高.5结语在目前的P2P视频点播系统上向大量的用户提供高清的视频点播服务是一个很大的挑战.本文通过分析中国科学技术大校园视频系统的日志记录,发现视频之间随用户的点播行为呈现出很强的社会网络现象.基于视频的社会网络特性,利用P2P视频点播系统已有的gossip协议,我们设计了一种新颖实用的基于社会网络的分布式视频推荐机制.仿真实验表明,我们提出的基于社会网络的分布式推荐可以给用户推荐相似的资源,同时极大缓解了带宽饥饿现象的发生.同时,对比其他策略,得出社会网络推荐机制可以有效地改进高清P2P视频点播系统的性能.References:[1]Jiang J W,Chan S H G,et al.1mbps p2p streaming:a global meas-urement study[R].Princeton University,May,2010.[2]Huang C,Li J,et al.Can internet video-on-demand be profitable?[C].In Proc.of ACM Sigcomm'07,Kyoto,Japan,Aug,2007.[3]Alessandria E,Gallo M,et al.P2P-tv systems under adverse network conditions:a measurement study[C].In Proc.of IEEE INFOCOM' 09,Rio de Janeiro,Brazil,Apr,2009.[4]Mao Jun-peng,Cui Yan-li,Huang Jian-hua,et al.Study of resource s load model of P2P network[J].Journal of Chinese Computer Sys-tems,2010,31,(2):215-219.[5]Shen Shi-jun,Li San-li.Novasky:design and implementation of a re-al world VoD/P2P system[J].Journal of Chinese Computer Sys-tems,2011,32(6):1041-1048.[6]He Y,Guan L.Solving streaming capacity problems in P2P vod sys-tems[J].IEEE Transactions on Circuits and Systems for Video Tech-nology,2010,20(11):1638-1642.[7]Wu W,Lui J C S,Ma R T B.Incentivizing upload capacity in P2P-vod systems:a game theoretic analysis[C].In Proc.of GameNets' 11,2011.[8]Yu H,Zheng D,et al.Understanding user behavior in large scale video-on-demand systems[C].In Proceedings of the EuroSys'06,Leuven,Belgium,Apr,2006.[9]USTC Video City[EB/OL].http://video.ustc.edu.cn,Feb,2010.[10]Bao Yi-ping,Yao Li,Zhang Wei-ming,et al.Model for computing feedback trustworthiness by recommending behavior in P2P networks [J].Journal of Chinese Computer Systems,2010,31,(11):2191-2195.[11]USTC Video Log[EB/OL].www.nhpcc.ustc.edu.cn/P2P/VoDDa-ta/,Feb,2010.[12]Fagiolo G.Clustering in complex directed networks[J].Physical Reviewe,2007,76(2):026107.[13]Tian Y,Wu D.A novel caching mechanism for peer-to-peer based media-on-demand streaming[J].Journal of System Architecture,2008,54(1-2):55-69.附中文参考文献:[4]毛军鹏,崔艳莉,黄建华,等.P2P网络资源负载模型研究[J].小型微型计算机系统,2010,31,(2):215-219.[5]沈时军,李三立.一个真实VoD/P2P系统Novasky的设计与实现[J].小型微型计算机系统,2011,32,(6):1041-1048.[10]鲍翊平,姚莉,张维明,等.对等网中一种面向推荐行为的反馈可信度评估模型[J].小型微型计算机系统,2010,31,(11):2191-2195.232小型微型计算机系统2013年。
面向动态拓扑网络的深度强化学习路由技术

doi:10.3969/j.issn.1001-893x.2021.06.001引用格式:伍元胜.面向动态拓扑网络的深度强化学习路由技术[J].电讯技术,2021,61(6):659-665.[WUYuansheng.Deepreinforcementlearn-ingroutingfordynamictopologynetworks[J].TelecommunicationEngineering,2021,61(6):659-665.]
面向动态拓扑网络的深度强化学习路由技术∗伍元胜∗∗(中国西南电子技术研究所,成都610036)
摘 要:针对现有智能路由技术无法适用于动态拓扑的不足,提出了一种面向动态拓扑的深度强化学习智能路由技术,通过使用图神经网络近似PPO(ProximalPolicyOptimization)强化学习算法中的
策略函数与值函数、策略函数输出所有链路的权值、基于链路权值计算最小成本路径的方法,实现了路由智能体对不同网络拓扑的泛化。仿真结果表明,所提方法可适应动态拓扑的变化并具有比传统的最短路由算法更高的网络吞吐量。关键词:智能通信网络;智能路由;深度强化学习;图神经网络;动态拓扑
开放科学(资源服务)标识码(OSID):微信扫描二维码听独家语音释文与作者在线交流享本刊专属服务
中图分类号:TN915 文献标志码:A 文章编号:1001-893X(2021)06-0659-07
DeepReinforcementLearningRoutingforDynamicTopologyNetworks
WUYuansheng(SouthwestChinaInstituteofElectronicTechnology,Chengdu610036,China)Abstract:Fortheshortcomingthatmoststate-of-the-artdeepreinforcementlearning(DRL)-basedrou-tingsolutionscannotbeusedinnetworkswithdynamictopology,aDRL-basedroutingsolutionfordynamictopologynetworksisproposed.Graphnetworksareusedtoapproximatepolicyfunctionandvaluefunctioninproximalpolicyoptimization(PPO)reinforcementlearningalgorithm,thelinkweightsareoutputbythegraphnetspolicyfunctionandtheleastweightpathiscomputedbythetraditionalconstrainedshortestpathroutingalgorithmbasedonthelinkweights,thusachievingthegeneralizationoftheroutingagenttodiffer-entnetworktopologies.Simulationresultsindicatethattheproposedroutingsolutioncanadapttodynamicnetworktopologyandoutperformstheshortestpathroutinginnetworkthroughputs.Keywords:intelligentcommunicationnetwork;intelligentrouting;deepreinforcementlearning;graphneu-ralnetwork;dynamictopology
Date Signed

AN EV ALUATION OF MULTIPLE PATH ROUTING AND ITS IMPACT ON QUALITY OF SER VICE IN MOBILE AD HOC NETWORKSbyGregory B.GerouREMOVEiA thesis submitted to the Faculty and the Board of Trustees of the Colorado School of Mines in partial fulfillment of the requirements for the degree of Master of Science(Mathematical and Computer Sciences).Golden,ColoradoDateGregory B.GerouApproved:ABSTRACTThe topic of multi-path routing protocols has been investigated previously;how-ever,the effectiveness of multi-path routing to improve quality of service by maximiz-ing bandwidth availability,as well as the criteria for determining multi-path routes, have not been clear.For that reason,we have developed a Proactive Disjoint Multi-path Routing protocol(PDMR)that we use to compare single-path routing versus multi-path routing.The results of this investigation illustrate the performance of multi-path routes and their impact on quality of service in a variety of mobile ad hoc network scenarios.iiiTABLE OF CONTENTSABSTRACT (iii)LIST OF FIGURES (vi)LIST OF TABLES (viii)ACKNOWLEDGEMENTS (ix)Chapter1INTRODUCTION (1)Chapter2RELATED WORK (5)2.1MANET Multi-path Routing Protocols (7)2.1.1Diffusing Algorithm for Shortest Multi-path(DASM) (8)2.1.2Disjoint Pathset Selection Protocol(DPSP) (9)2.1.3Multi-path Dynamic Source Routing(MDSR) (11)2.1.4The Graph Multi-path Routing Protocol(GMR) (13)2.1.5The Split Multi-path Routing Protocol(SMR) (15)2.2MANET QoS Frameworks (16)2.2.1Core-Extraction Distributed Ad Hoc Routing(CEDAR) (16)2.2.2Insignia (19)2.2.3Stateless Wireless Ad Hoc Networks(SWAN) (19)Chapter3DISJOINT PATH CLASSIFICATION (21)Chapter4PDMR (26)4.1Network Topology Dissemination (27)4.2Multi-path Discovery Algorithm (28)4.3Multi-path Use Policies (31)4.4The Role of SWAN (32)Chapter5SIMULATION RESULTS (33)5.1Environment (33)5.2Static Network Models (37)5.3Mobile Network Models (46)5.4Contrived Network Model (53)ivChapter6CONCLUSIONS AND FUTURE WORK (58)6.1Conclusions (59)6.2Future Work (60)REFERENCES (63)vLIST OF FIGURES2.1DASM’s path classification example (10)2.2MDSR-Protocol1example (13)2.3MDSR-Protocol2example (13)2.4A comparison of GMR and MDSR (15)2.5CEDAR example network (18)3.1Node disjoint route with two paths (22)3.2Strict link disjoint route with two paths (23)3.3Loose link disjoint route with two paths (23)3.4Semi-node disjoint route with two paths of equal length (24)3.5Semi-node disjoint route with two paths of different length (24)3.6Disjoint route classification hierarchy (25)5.1Route frequencies for various network types(including Route Not Founderror frequencies)as the number of nodes in the network increases..38 5.2Route frequencies for various network types(without Route Not Founderror frequencies)as the number of nodes in the network increases..39 5.3Average route lengths for networks simulated within a region of300mx600m,as the number of nodes in the network increases (40)5.4The average difference between the two paths included in each multi-path route,as the number of nodes in the network increases (42)5.5Total delivery rate for a static network with seven concurrent QoStraffic sessions as the number of simultaneous best effort traffic sessionsvaries (43)5.6Best effort delivery rate for a static network with seven concurrent QoStraffic sessions as the number of simultaneous best effort traffic sessionsvaries (44)vi5.7QoS delivery rate for a static network with seven concurrent QoS trafficsessions as the number of simultaneous best effort traffic sessions varies45 5.8QoS delivery rate for a static network with four concurrent QoS trafficsessions as the number of simultaneous best effort traffic sessions varies46 5.9QoS delivery rate for a static network with one QoS traffic stream asthe number of simultaneous best effort traffic sessions varies (47)5.10QoS delivery rate for a static network with25best effort traffic sessionsas the number of simultaneous QoS traffic sessions varies (48)5.11QoS delivery rate for a static network with30best effort traffic sessionsas the number of simultaneous QoS traffic sessions varies (48)5.12Total delivery rate for a mobile network with seven concurrent QoStraffic sessions as the number of simultaneous best effort traffic sessionsvaries (49)5.13QoS delivery rate for a mobile network with seven concurrent QoStraffic sessions as the number of simultaneous best effort traffic sessionsvaries (49)5.14QoS delivery rate for a mobile network with two concurrent QoS trafficsessions,two concurrent best effort traffic sessions,and variousfixedroute lengths (50)5.15QoS delivery rate for a mobile network withfive concurrent QoS trafficsessions,five concurrent best effort traffic sessions,and variousfixedroute lengths (51)5.16QoS delivery rate for a mobile network with ten concurrent QoS trafficsessions,ten concurrent best effort traffic sessions,and variousfixedroute lengths (52)5.17Network topology for the contrived scenario (54)5.18Delivery rates for the contrived scenario (56)5.19Throughput rates for the contrived scenario (57)viiLIST OF TABLES2.1A breakdown of the fundamental components of each of the discussedmulti-path routing protocols (17)3.1Relationships between disjoint route types and existing multi-path pro-tocols (25)4.1Pseudocode describing the PDMR route discovery algorithm (30)5.1Simulation parameters (34)5.2Network model parameters for static(0m/s)simulations (36)5.3Network model parameters for mobile(1m/s)simulations (36)5.4Simulation parameters (54)viiiACKNOWLEDGEMENTSI am happy to thank my adviser,Dr.Tracy Camp,for her invaluable guidance and unending patience.I would also like to thank my thesis committee members,Dr. Mike Colagrosso and Dr.Jason Liu.Through the course of my research,their input has always steered me in the right direction.I also appreciate the support I have received from members of the Toilers research group.ixChapter1INTRODUCTIONA mobile ad hoc network(MANET)is a network of mobile nodes capable of com-municating without the use of any static network infrastructure.Nodes comprising a MANET are not always capable of communicating directly with the entire network. Limited transmission range implies multi-hop routing,the ability for a given node to transmit data to a destination by passing a packet through intermediate nodes. Transmission range limitations also introduce the hidden node problem,the poten-tial for transmission collisions observed by a node with two transmitting neighbors, each unaware of the other’s transmission.Multiple-hop routing and the hidden node problem are two examples of difficulties created by the wireless and mobile properties of MANETs.As applications from traditional wired networks are moved into the mobile wireless environment,these properties reveal new problems to be solved.Quality of Service(QoS)is an example of a traditional wired network desired property that creates new problems when applied to mobile ad hoc networks.QoS refers to the potential performance requirements of network traffic.For example,a particular application may demand that traffic be delivered with a quantified min-imum delay,or be allowed a specified amount of network bandwidth.Relative to wired networks,reliable QoS is difficult to achieve in MANETs due to higher network volatility.This volatility leads to difficulty in measuring network state,and applying those measurements to QoS traffic routing.Stine and Veciana suggest that to include useful quality of service mechanisms, MANET protocols and frameworks should place a greater emphasis on node statethan on link state[1].In terms of quality of service,node state refers to the available network resources and known characteristics observed at a given node,while link state refers to those resources and characteristics observed over a logical communication link between two nodes.This link is defined by the ability of neighboring nodes to directly communicate.An example of a link state measurement might be node A is transmitting1kb/s to node B;an example of a node state measurement might be node A is observing a local bandwidth usage of3.2kb/s(created both by its own transmissions and those of its neighbors).The emphasis of node state over link state impacts MANET routing protocols because it includes the impact of neighbor traffic on bandwidth availability.Addition-ally,it prioritizes efficient use of the RF transmission space surrounding a node rather than that of a link between a pair of nodes.As described in the previous example, link state measurements do not necessarily indicate available network resources.For example,suppose neighbor nodes A and C are reaching their maximum receivable bandwidth,but are not sending or receiving packets to or from each other.Although neither A nor C observe any bandwidth on their shared link,given that neither is capable of receiving further traffic,any new traffic between the two nodes will likely result in a transmission collision.In this example,a QoS framework based on link state measurements would likely yield poorer performance than a QoS framework based on node state measurements.The change in approach from a link-based to node-based paradigm is also evi-dent in comparing earlier QoS work to current QoS work.For example,the Core-Extraction Distributed Ad hoc Routing Algorithm(CEDAR)[2]and an in-band signaling system,Insignia[3],use link-based metrics while a later approach such as SWAN[4,5]relies on node-based measurements.SWAN has proved it is possible toobtain good quality of service by using a stateless,reservation-less approach.Specif-ically,SWAN enables the monitoring and shaping of available bandwidth.Although robust approaches to QoS in MANETs have been developed and pub-lished,most rely on the existence of a single-path that is capable of satisfying the resource requirements for the dataflow.A single path traffic session incurs a variety of costs such as large bandwidth use due to network traffic along the single path,in-creased delay due to possible transmission collisions and saturated forwarding queues, reduced battery life due to requisite retransmissions,etc.When a single path is used to transmit a stream of traffic,the costs associated along the single path with that traffic are imposed on the nodes along that path.This problem has motivated the development of protocols that leverage multi-paths to distribute the bandwidth costs associated with a data stream over a broader range of the network.In order for multiple paths to benefit traffic delivery,the following requirements must be met:1.The source and destination nodes are not bandwidth constrained.2.Intermediate nodes on the multi-paths are under load such that a)no singlepath can satisfy the bandwidth requirement,and b)considered together,their bandwidth capacity is sufficient to satisfy the bandwidth requirement.The goals of this thesis are twofold.Thefirst goal is to present our new multi-path protocol,the Proactive Disjoint Multi-path Routing protocol.The second goal is to use PDMR as a tool for evaluating single-versus multi-path routing techniques.Although the above requirements for improving traffic delivery with multi-path routing are relatively straightforward,it is necessary to determine if their frequency is sufficient to observe a significant improvement with multi-path routing relative to single-path routing.Chapter2covers related work,including both existing multi-path routing pro-tocols and established QoS frameworks.Chapter3covers the topic of disjoint routes, and the criteria for classifying various disjoint path types.This analysis is then ap-plied to the previously covered multi-path routing protocols.Chapter4describes the details of our Proactive Disjoint Multi-path Routing protocol(PDMR):a topology dissemination component,a multi-path discovery algorithm,path use policies,and admission and rate control components.Chapter5presents our PDMR simulation results and analysis.Chapter6presents the conclusions we draw from the results obtained over the course of our research.Chapter2RELATED WORKMultiple path routing is a concept that has been applied to several applications [16].However,the fundamental difficulties presented in MANETs caused by node mobility and communication over a volatile medium make new study of multi-path routing protocols within this context interesting.A significant amount of work has been published on the subject of multi-path routing in MANETs.In addition to the work covered in detail later in this chapter,Villela and Duarte present analysis regard-ing traffic throughput maximization using multi-path routing[9].Chen,Druschel and Subramania present a multi-path forwarding policy[10].Cidon,Rom,and Shavitt develop a mathematical model of a multi-path protocol with a QoS reservation mech-anism that shows1)in most networks more than two or three disjoint paths offer no improvement in performance,and2)real network behavior may deviate from the conditions required to obtain a benefit in performance from multi-path routing[11]. Ogier,Rutenburg,and Shacham present algorithms for computing disjoint paths, without an emphasis on wireless environments[12].Sidhu,Nair,and Abdallah also present a network-generic algorithm for disjoint path discovery[13].Taft-Plotkin, Bellur,and Ogier suggest multi-path routing as an implementation for providing QoS services[14].Vutukury,and Garcia-Luna-Aceves suggest a protocol that calculates multi-paths through the use of distance vector routing.Much of this work includes protocols for discovering and using multi-paths;additionally,Corson and Macher of-fer an evaluation framework for multi-path protocols[23].The framework identifies the following fundamental components of a multi-path protocol:1.Multiple Route Discovery.The procedure by which the protocol selects multi-path routes.This procedure generally includes provisions to avoid path looping, and heuristics to generate disjoint multi-path sets.2.Filtering Provision.A protocol component capable of eliminating undesirablemulti-path sets based on a metric such as path length,available bandwidth,etc.3.Path Usage Policy.The policy by which the protocol decides what and whento transmit along each of the paths in the multi-path set.Examples of such a policy are:•Load Balancing.The transmitting node will send each packet along the least recently used path.For instance,if the multi-path set is comprised of two paths,the transmitting node will alternate packet transmissions between the two paths.•Redundancy.The transmitting node will transmit each packet along all paths.In the case of redundant path usage,the source node must include a policy to determine the path order for the duplicate packet transmissions.When a node must transmit the same packet to two different destinations individually,a data forwarding policy defines the order of the transmissions to each of thesedestinations.Examples of a data forwarding mechanism include a round-robin approach,or a heuristic such as path length to determine path priority.The need for path prioritization in a forwarding mechanism is motivated by the dif-ference in predictable behavior between path options.4.Multi-path Maintenance Heuristic.In single-path routing,a set of rules mustbe built into the routing protocol to enable path recomputation.The only ad-ditional complexity,compared to single path routing,is the inclusion of multi-paths.Traditional single-path maintenance heuristics must be updated or re-placed to enable efficient use of multi-paths.5.Underlying Single Path Routing Protocol.The use of multi-paths implies theavailability of multi-paths,and multi-paths are not always available.If no multi-path is available,but a single path is available,the multi-path protocol must be capable of using and maintaining that single path.These components comprise a framework for evaluating multi-path routing proto-cols.In Section2.1,we discuss the prominent multi-path routing protocols currently available.We summarize these protocols by categorizing them via these fundamental components.Section2.2then covers the significant published QoS frameworks for MANETs.2.1MANET Multi-path Routing Protocols2.1.1Diffusing Algorithm for Shortest Multi-path(DASM)When a next hop is not defined for a given destination at a given node,that node queries its neighbors for information regarding the destination.If any of the neighbors have the requested destination node in their routing tables,a reply is sent to the querying node indicating the next hop and the total distance to the destination. If no neighbor has the requested destination node in their routing tables,the request is repeated.This process continues until a querying node has been answered with a distance to the destination.At network startup,when all routing tables are empty, this process takes the form of aflooded route request.As replies are sent back to the source,neighboring nodes promiscuously update their routing tables.We note that neighbors of each querying node may not hear route reply messages sent to the querying node.Thus,although the querying node may have successfully discovered the next hop to the destination,the route request will continue to be forwarded.Route replies include a hop distance measurement,enabling querying nodes to maintain a shortest path hop,and additionally maintain the shortest multi-path hop.DASM’s path usage policy dictates that when the shortest path fails,a shortest multi-path is used if one is available.If no shortest multi-path route is available when the shortest path route fails,a new route request is triggered.If a shortest multi-path is available and if both the shortest path and the shortest multi-paths have failed,a new route discovery is triggered.Table2.1describes DASM according to the fundamental components of multi-path protocols,and compares it to the other multi-path protocols described in this chapter.2.1.2Disjoint Pathset Selection Protocol(DPSP)Fig.2.1.DASM’s path classification examplepath information sufficient to build(at least)a partial graph representation of the network between the source and destination nodes.DPSP introduces the notion of path reliability for use in path selection.The protocol defines reliability as a function of both a path’s length and the delivery rate over each of the path’s links.This function is essentially DPSP’s pathfiltering provision.A path set is generated by performing a search on the graph for a maximally reliable path(according to DPSP’s reliability function).This path is added to a multi-path set;the intermediate nodes in the discovered path are removed from the graph(which we term a node disjoint filter),and a new search is performed.The algorithm includes a threshold mechanism to determine when additional paths no longer offer a reliability benefit,according to DPSP’s reliability function.Once a multi-path set has been calculated,data is transmitted redundantly on each of the paths,theoretically improving the probability the packetized data will arrive at the destination.The data forwarding mechanismuses a round robin scheme to determine a path order to be used for redundant datapacket transmissions.If the reliability of a calculated route falls below a defined threshold,a new route calculation is triggered.It is possible for the protocol to calculate a single path route,given that it meets the requirements to be considered reliable.Table2.1describes DPSP according to its fundamental components,and compares it to the other multi-path protocols described in this chapter.2.1.3Multi-path Dynamic Source Routing(MDSR)requests are forwarded at most once by any given node involved in any given route discovery.These additional routes are generally longer,as these route requests arrive after the primary source route’s request arrives.Route replies for each route are sent to the source.The MDSR route usage policy dictates secondary routes are only used after the primary route has failed.The use of previously discovered secondary routes reduces route discovery overhead.If only one RREQ is received by the destination (i.e.,there is only one path from source to destination),MDSR acts essentially the same as the single path DSR protocol.MDSR offers two mutually exclusive multi-path maintenance heuristics.The authors label these versions of MDSR as“Protocol1”and“Protocol2”.Protocol 1prevents intermediate node route caching for alternate routes,which implies that only the source is capable offixing a broken path.Figure2.2depicts an example of Protocol1for a given source,S,and destination,D;the dotted lines represent the primary route,and the solid lines represent the alternate paths.Protocol2,on the other hand,allows intermediate nodes to cache alternate routes;intermediate nodes use this information saved tofix routes when broken links are detected.For instance, if an intermediate node determines a current route has failed,it begins using any available cached alternate route to the destination.If no such route is cached,a route error is sent to the next upstream node where the process is repeated.Figure 2.3depicts an example of Protocol2;the dotted lines represent the primary route, and the solid lines represent the alternate routes.Protocol1offers the simplicity of maintaining alternate disjoint paths for each traffic session only at the source;Protocol 2requires more overhead for maintaining alternate disjoint paths at all intermediate nodes,but is able to offer backup path maintenance.Table2.1describes MDSR according to its fundamental components,and compares it to the other multi-pathprotocols described in this chapter.Fig.2.2.MDSR-Protocol1exampleFig.2.3.MDSR-Protocol2example2.1.4The Graph Multi-path Routing Protocol(GMR)from all received route requests are combined and forwarded as a single route request. This method enables the destination to build a graph representation of the network and to determine multiple link disjoint routes.Once these paths are communicated to the source,data packets begin transmitting.The published work on GMR does not specify what policy is used to determine how the multi-paths are employed,but it is reasonable to assume that once it has been determined that all calculated paths have failed,a new route discovery process is triggered.Additionally,it is possible for GMR to function as a single path routing protocol,assuming only a single path exists between the source and destination nodes.The motivation for GMR is a limitation of MDSR.Duplicate route requests are ignored in MDSR,just as they are in DSR.Thus,the set of multi-path routes discovered in MDSR is defined by the sequence in which route requests arrive at each node.For instance,in MDSR,a node that receives a route request will forward it and, therefore,belong to the growing potential route included in that request.All other route requests received by that node are ignored,which implies the route requests that reach the destination are comprised of mutually exclusive node sets.GMR,on the other hand,incorporates duplicate route requests,thus allowing potential overlapping route requests to reach the destination.Equipped with these additional,potentially overlapping options,the destination is more capable of calculating disjoint paths. Figure2.4illustrates this idea.The solid connecting lines indicate links potentially discovered by MDSR.The combination of the solid connecting lines and the dashed connecting lines are links potentially discovered by GMR.Figure2.4illustrates that GMR is capable of providing enough topological data to the destination node to build a graph representation of the pertinent network subset.MDSR is not capable of building such a graph.Table2.1describes GMR according to its fundamentalcomponents,and compares it to the other multi-path protocols described in this chapter.Fig.2.4.A comparison of GMR and MDSR2.1.5The Split Multi-path Routing Protocol(SMR)either1)continue using the single available path or2)trigger a new route discovery.Obviously the advantage of not automatically triggering a new route request is a decrease in routing overhead traffic.However,SMR’s authors also acknowledge that waiting until all available paths to a destination have failed implies that no data can be sent to the destination during the subsequent route discovery.SMR is configurable such that nodes initiate new route discoveries as soon as the shortest route breaks to ensure that an alternate path is available for data transmission during the route discovery process.If only one path is available,regardless of how SMR is configured, a new route discovery is triggered upon the failure of this single path.Table2.1 describes SMR according to its fundamental components,and compares it to the other multi-path protocols described in this chapter.2.2MANET QoS FrameworksCEDAR,the Core Extraction Distributed Ad hoc Routing algorithm[2]is a proposed QoS routing framework for small-to medium-sized ad hoc networks(tens to hundreds of nodes).The nodes in the network dynamically elect an approximate minimum dominating set of core nodes.A set of core nodes is considered a minimum dominating set if they represent the minimum number of nodes required to span thenetwork,i.e.,every non-core node is a one-hop neighbor of at least one core node. This network of core nodes is used to propagate link bandwidth information in two modes.First,information regarding a significant increase in available bandwidth on a given link is propagated by an increase wave,i.e.,a packet that is repeated over a broad range of the core network.A significant decrease in available bandwidth on a given link is propagated by a decrease wave,i.e.,a packet that is repeated over a smaller range of the core network.Essentially,information about usable links is shared over a large amount of the network and information about constrained links is shared locally.Figure2.5illustrates an example CEDAR network.The nodes depicted in solid black comprise the network’s core or the network’s approximated minimum dominating set;all other nodes are depicted as white-centered circles.The solid connecting lines denote links between pairs of nodes,while the dotted lines indicate the core network.Fig.2.5.CEDAR example network2.2.2InsigniaThe Stateless Wireless Ad Hoc Networks(SWAN)protocol[4]was also developed by the author’s of Insignia[3].SWAN offers a stateless/reservation-less efficient and effective QoS framework.The two components in SWAN are a rate controller and an admission controller.The goal of SWAN is to adapt bandwidth usage according to the measured one hop delay to ensure that1)delay does not become excessive and2) network bandwidth does not become saturated to the point of significant packet loss and reduced throughput.SWAN improves upon INSIGNIA by abandoning the idea of bandwidth reservations in favor of efficient use of the RF medium on a node-by-node basis.This modification improves network efficiency.SWAN handles best effort(non-priority)traffic easily,and includes provisions for priority packets to enable qualityof service support.QoS packets are given priority treatment in terms of admission and bandwidth usage.Upon network saturation,best effort packet throughput and delivery rates suffer before high priority throughput and delivery rates are impacted. In other words,by sacrificing best effort traffic delivery,a high network throughput of high priority traffic is maintained.In non-saturated conditions,high priority traffic continues to receive queuing preference,minimizing delay for QoS packets.。
指导老师舒炎泰教授

Computer Department
Network Layer-Open Research Issues
Scalability
分层路由协议由于其管理的复杂性和困难可能只有部分解决了这个 问题。地理路由依赖于GPS的存在或类似定位的技术,增加了WMNs成 本和复杂性。因而需要开发新的可扩展路由协议。
Tianjin University Computer Department
Design Approaches of a Single-channel MAC protocol
Modifying Existing MAC Protocols
例如,在一个802.11 mesh网中, MAC层协议可以通过调整CSMA/CA的参 数得到改善(如CW的大小,修改backoff程序)。但该方法仅能实现低的 端到端的吞吐率,因为它不能大量的减少邻居节点间竞争的概率。
Tianjin University
Computer Department
PHY- MIMS
M,N,K,L的值不同会产生各种多天线系统;
接收方有多个天线接受,发送方是单个通道 (K=1,M=1以及或 L>1或N>1)的设计已经被建议,接受方能够收到至少一个好信 号的概率是很高的。
Tianjin University Computer Department
Tianjin University Computer Department
Overview of Research Topics
Physical Layer MAC Layer Network Layer Transport Layer
Application Layer
多目标MIN-MAX度最小树问题及其求解

多目标MIN-MAX度最小树问题及其求解魏欣;马良【摘要】在多目标最小生成树问题和MIN-MAX度最小树问题的基础上,探讨使生成树最大顶点度数以及总权重都尽可能小的另类多目标MIN-MAX度最小生成树问题.分析了这一特殊的顶点度约束与Hamilton路的关联性质,在此基础上设计了先Hamilton路再MIN-MAX度最小树的独特求解方案.根据初始条件不同,当网络图不存在Hamilton路时,引入改进的蚁群优化算法,将转移概率由基本的指数形式改进为线性形式,在不影响求解质量的前提下,提高计算效率.针对以上策略,设计了相应的求解方案,并在计算机上用Delphi编程实现.大量数值算例验证表明,算法能快速有效地求解多目标情形下的MIN-MAX度最小生成树问题.【期刊名称】《上海理工大学学报》【年(卷),期】2019(041)003【总页数】5页(P231-235)【关键词】多目标;MIN-MAX度;生成树;Hamilton路【作者】魏欣;马良【作者单位】上海理工大学管理学院,上海 200093;上海理工大学管理学院,上海200093【正文语种】中文【中图分类】O224最小生成树(minimum spanning tree, MST)问题是运筹学、网络优化中的一个基本而又经典的优化问题[1],早在20世纪50年代就已圆满解决,在生活中有着大量的实际应用。
此后,在此基础上,又逐步延伸出一系列的扩展问题,如:度约束最小生成树问题、多目标最小树问题、MINMAX度生成树问题、MIN-MAX度最小树问题等[2-8]。
由于这些扩展问题在求解难度上与经典的最小树问题截然不同,目前都已归入所谓的NP难题范畴。
其中,对生成树中各节点的度数有某种要求或限制,以及网络图本身就带有多个权重属性的多目标问题,在现实社会经济生活中经常会遇到,具有重要的应用价值。
例如:集成电路布线中往往对相关节点连出的线数有某种工艺限制[9-10],这就是典型的度约束最小树问题。
计算机网络自顶向下方法第七章中文版答案全文优选

WRI 研究生0601最新精选全文完整版(可编辑修改)7复习题流式存储音频/视频: 暂停/恢复, 重新定位, 快进, 实时交互的音频视频: 人们进行实时的通信和响应。
第一阵营:TCP/IP协议中的基本原理没有变化, 按照需求增加带宽, 仍然使用有超高速缓存, 内容分布, 多点覆盖的网络。
1.第二阵营: 提供一个可以使应用在网络中节省带宽的网络服务。
2.第三阵营: 有区别的服务: 提出了在网络边缘的简单分类和维护方案。
并且根据他们在路由队列中的级别给出了不同的数据包和级别。
6.1: 简单, 不需要meta file 和流媒体服务器3. 6.2:允许媒体播放器通过web服务器直接进行交互, 不需要流媒体服务器。
4. 6.3:媒体播放器直接与流媒体服务器进行交互, 以满足一些特殊的流媒体应用。
5.端到端时延是分组从源经过网络到目的地所需要的时间。
分组时延抖动是这个分组与下个分组的端到端时延的波动。
6.在预定的播放时间之后收到的分组不能被播放, 所以, 从应用的观点来看, 这个分组可以认为是丢失了。
7.第一种方案: 在每n个数据块之后发送一个冗余的数据块, 这个冗余的被。
【the redundant chunk is obtained byexclusive OR-ing the n original chunks】8.第二种方案:随起始的数据流发送一个低分辨率, 低比特率的方案, 这种交错不会增加数据流的带宽需求。
不同会话中的RTP流: 不同的多播地址同一会话中不同流: SSRC fieldRTP和RTCP分组通过端口号区别。
9.传达报告分组: 包括分组丢失部分的信息, 最后序列号, 两次到达时间间隔的抖动。
发送报告分组: 大部分目前产生的RTP分组的时间戳和wall clock time , 发送分组的数量, 发送字节的数量, 源描述分组: 发送方的e-mail 地址, 发送方的姓名, 产生RTP流的应用。
基于BATMAN-adv的多跳无线网络

第19卷 第3期太赫兹科学与电子信息学报Vo1.19,No.3 2021年6月 Journal of Terahertz Science and Electronic Information Technology Jun.,2021文章编号:2095-4980(2021)03-0433-05基于BATMAN-adv的多跳无线网络赵志豪1,陈娅冰1,张戬 2(1.烟台工程职业技术学院信息与传媒系,山东烟台 264006;2.哈尔滨工业大学计算机科学与技术学院,黑龙江哈尔滨 150001)摘 要:无线自组网具有去中心化、多信道、多跳等特点,适用于专网和临时应急网络的快速搭建。
为了使网络具有更宽的覆盖、更强的绕射能力和更高的吞吐量,系统选择一种支持IEEE802.11 ac的三射频接口物理层节点,在此基础上,结合BATMAN-adv路由协议进行研究,实现了一种基于多接口多信道的自组网系统。
仿真实验数据表明,该系统在5G频谱中能够全双工使用正交信道降低干扰,将网络链路拓展至5跳以上,网络损耗衰减低于30%,网络吞吐量高于300Mbps。
关键词:自组网;多信道;正交信道;吞吐量中图分类号:TN915.03 文献标志码:A doi:10.11805/TKYDA2019571Multi-hop wireless network research based on BATMAN-advZHAO Zhihao1,CHEN Yabing1,ZHANG Jian2(rmation and Media Department, Yantai Engineering & Technology College, Yantai Shandong 264006, China;2.College of ComputerScience and Technology, Harbin Institute of Technology, Harbin Heilongjiang 150001, China)Abstract:Self-organized wireless network, featured by non-centralized, multi-channel, multi-hop, meets the requirements of emergence networks and dedicated networks. To gain the network performancewith longer range cover, non-visible communication and higher throughput, tri-radio nodes supportingIEEE802.11 ac protocol are selected. Combining with Better Approach To Ad-hoc NetworkAdvanced(BATMAN-adv) routing protocol, a system of self-organized wireless network with multi-channeland multi-interface is realized. The field test data indicates that the system extends coverage to 5 hops with30%throughput loss and the throughput is up to 300Mbps with orthogonal channels under full duplexmode in 5G spectrum.Keywords:self-organized;multi-channel;orthogonal channels;throughput无线网状网络(Wireless Mesh Network,WMN)[1]是一种区别于传统无线网络的新型通信网络。
BitTorrent 基于临近原则

基于邻近原则的BitTorrent实验研究张增斌,陈阳*,邓北星,李星(清华大学电子工程系,北京,100084)摘要:BitTorrent(BT)是一种基于P2P的文件共享软件,有着十分广泛的应用。
在BT中,默认参与节点随机选择其它节点作为网络中的邻居构成覆盖网络,不能根据节点的位置优化覆盖网络,影响了BT中文件传输的性能。
BT的有偏邻居选择,指的是BT Tracker参照BT Client在互联网中的位置,向BT有针对性地提供相应的邻居,优化BT的覆盖网络,使得BT的文件传输效率得到提高。
本文提出了一种基于邻近原则优化BT文件传输速率的机制。
首先利用网络坐标对BT网络参与节点在互联网中的位置进行计算;之后,根据网络坐标,对于BT参与节点的邻居进行了有偏选择;BT 参与节点经过分布式聚类得到基于邻近原则的覆盖网络。
仿真实验显示,基于网络坐标的有偏邻居选择对于BT整体文件传输性能有较大的提升。
关键词:BitTorrent;网络坐标;有偏邻居选择;邻近原则近些年来,基于P2P的各种网络应用越来越广泛,其中影响最大的应用是文件共享。
P2P文件共享以BitComet、eMule、Azureus等BitTorrent(BT)[1]类软件为代表。
BT流量在整个网络流量中比例非常高,2004的统计数据就已表明,BT流量已经占据了所有P2P流量的53%,而P2P的流量占整个网络流量的80%以上[2],最近几年更是飞速增长,BT流量已经升至70%-80%。
如何更为有效的提高BT文件传输的效率,对于提高网络服务的性能、降低网络的负载都有极为重要的现实意义。
网络坐标是近年来出现的一种通过少量的端到端测量来预测网络距离的工具,其理论基础是网络中节点间的距离(延迟等)大部分满足三角不等式,因此可以根据极少量的测量结果将节点映射到欧几里德空间中的一个点上,从而根据任意两个点的坐标,就可以估算出他们之间的距离。
将网络坐标应用于BT中,从而使每个节点能够根据坐标的信息选择较近的节点作为邻居,得到优化的覆盖网络,大幅度提高性能。
- 1、下载文档前请自行甄别文档内容的完整性,平台不提供额外的编辑、内容补充、找答案等附加服务。
- 2、"仅部分预览"的文档,不可在线预览部分如存在完整性等问题,可反馈申请退款(可完整预览的文档不适用该条件!)。
- 3、如文档侵犯您的权益,请联系客服反馈,我们会尽快为您处理(人工客服工作时间:9:00-18:30)。
1580IEEETRANSACTIONSONMULTIMEDIA,VOL.9,NO.8,DECEMBER2007OnMaximizingTreeBandwidthforTopology-AwarePeer-to-PeerStreamingXingJin,StudentMember,IEEE,W.-P.KenYiu,StudentMember,IEEE,S.-H.GaryChan,SeniorMember,IEEE,andYajunWang
Abstract—Inrecentyears,therehasbeenanincreasinginterestinpeer-to-peer(P2P)multimediastreaming.Inthispaper,wecon-siderconstructingahigh-bandwidthoverlaytreeforstreamingser-vices.Weobservethatunderlayinformationsuchaslinkconnec-tivityandlinkbandwidthisimportantintreeconstruction,becausetwoseeminglydisjointoverlaypathsmaysharecommonlinksontheunderlay.Wehencestudyhowtoconstructahigh-bandwidthoverlaytreegiventheunderlaytopology.WeformulatetheproblemasbuildingaMaximumBandwidthMulticastTree(MBMT)oraMinimumStressMulticastTree(MSMT),dependingonwhetherlinkbandwidthisavailableornot.WeprovethatbothproblemsareNP-hardandarenotap-proximablewithinafactorof(23+),forany0,unless=NP.Wethenpresentapproximationalgorithmstoaddressthemandanalyzethealgorithmperformance.Furthermore,wediscusssomepracticalissues(e.g.,groupdynamics,resilienceandscalability)insystemimplementation.WeevaluateouralgorithmsonInternet-liketopologies.Theresultsshowthatouralgorithmscanachievehightreebandwidthandlowlinkstresswithlowpenaltyinend-to-enddelay.MeasurementstudybasedonPlan-etLabfurtherconfirmsthis.Ourstudyshowsthattheknowledgeofunderlayisimportantforconstructingefficientoverlaytrees.
IndexTerms—Overlaytree,peer-to-peerstreaming,topology-aware,treebandwidth.
I.INTRODUCTION
WITHthepopularityofbroadbandInternetaccess,there
hasbeenanincreasinginterestinmediastreamingservices.Recently,P2Pstreaminghasbeenproposedanddevelopedtoovercomelimitationsintraditionalserver-basedstreaming.InaP2Pstreamingsystem,cooperativepeersself-organizethemselvesintoanoverlaynetworkviaunicastconnections.Theycacheandrelaydataforeachother,therebyeliminatingtheneedforpowerfulserversfromthesystem.Inthispaper,weconsiderbuildingahigh-bandwidthoverlaytreeforsteaming.Astreamingvideousuallyhasacertainbi-trate.Toensuregoodstreamingqualityatahost,theincomingbandwidthofthehostshouldbehigherthanorequaltothestreamingbitrate.Wehenceneedtobuildatreewithenoughtreebandwidthtosupportthestreaming.Heretreebandwidth
ManuscriptreceivedOctober31,2006;revisedJuly29,2007.ThisworkwassupportedinpartbytheResearchGrantCounciloftheHongKongSpecialAdministrativeRegion(HKUST611107)andHongKongInnovationandTechnologyCommission(GHP/045/05).TheassociateeditorcoordinatingthereviewofthismanuscriptandapprovingitforpublicationwasProf.KlaraNahrstedt.TheauthorsarewiththeDepartmentofComputerScienceandEngineering,HongKongUniversityofScienceandTechnology,Kowloon,HongKong,China(e-mail:csvenus@cse.ust.hk;kenyiu@cse.ust.hk;gchan@cse.ust.hk;yalding@cse.ust.hk).DigitalObjectIdentifier10.1109/TMM.2007.907459
Fig.1.Theimpactofunderlayinformationtooverlaytreeconstruction.(a)Naiveconstructionofanoverlaytree(thelabelalongahostisitsdegreebound).(b)Treebandwidth=0:5
(thelabelalongalinkistheresidual
bandwidthofthelink).(c)Atreewithhighertreebandwidthof1.0.
isdefinedastheminimumpath-bandwidthofanoverlaytree[1].Notethataverageincomingbandwidthofhostsmayalsobeusedtoevaluateatree.However,whentheincomingbandwidthofahostislowerthanthestreamingbitrate,thehostwillen-counterpacketlossandthereceivedstreamingqualitywillbereduced.Averageincomingbandwidthcannotindicatethepor-tionofhostswithgood(orbad)streamingqualities.Wehencefocusontreebandwidthinsteadofaverageincomingbandwidthinthispaper.Somepreviousworkimposesdegreeboundsonhosts.Thatis,thedegreeofahostinanoverlaytreecannotexceedacertainbound[2],[3].Thisapproachcaneffectivelyreducecongestionnearhosts.However,itcannotimprovetreebandwidthifthebandwidthbottlenecksoccuratintermediatelinksinpathsin-steadofedgelinksnearhosts.Furthermore,twoseeminglydis-jointoverlaypathsmaysharecommonunderlaylinks;thereforetheselectionofoverlaypathswithouttheknowledgeofunderlaymayleadtoseriouslinkcongestion.Notethatinthispaper,alinkreferstoaphysicalunderlayconnectionbetweentworoutersorbetweenarouterandahost;whileapathreferstoanend-to-endconnectionbetweentwohosts,whichmaycontainmultipleun-derlaylinks.Furthermore,thebandwidthoflinksorpathsreferstotheresidualbandwidthalonglinksorpaths.WeshowatreeconstructionexampleinFig.1.,andarefourhosts,whosedegreeboundsare2,2,2and1,respectively.Rectangleswithlabels1to5arerouters.Withoutanyknowledgeoftheun-derlay,atreemaybeconstructedasFig.1(a)shows.SupposethatthelinkconnectivityandlinkbandwidthontheunderlayisasshowninFig.1(b).Wecanseethatthistreeisofbandwidth0.5(thebottlenecklinkisbetweenrouters3and4).Giventheunderlayinformation,wecaninfactbuildatreewithhighertreebandwidthof1asshowninFig.1(c).Therefore,underlayinfor-mationisimportantforbuildingahigh-bandwidthoverlaytree.