Analysis of a hybrid arch for cost-effective streaming media distribution.

By D. Xu et al.

This paper proposed a hybrid system which combines CDN and P2P streaming. In addition a quantitative model is proposed to analyze the CDN-p2p system: includes hand-off time, remaining requesting peers, total committed p2p streaming capacity and instantaneous p2p streaming capacity(??).

This system is cost-effective.

shortcoming:
Why need to combine p2p and cdn? what is the shortcoming of P2P streaming system?
The hand-off modeling is not incorperated into the protocol design? Some detail of the model is not described and some equation need to be proved.

Promise: using collectcast

By M. Hefeeda
This paper proposed a novel p2p streaming system and it incorporated the dynamic adaptation, peer lookup and packet loss.
The system considers the scenario that several peers provides video segment to one receiver which is different with other system and this method tries to increase the best video quality that the receiver can receive.
In this paper, the status of the peers are monitored and the roles (active or non-active) are changed dynamically.
Three selections algs are proposed and compared: random, end-to-end, topology-aware selection.

Each peer is associated with two parameters: declared rate and available rate.

Shortcoming:
How to coordinate the sending of packers from several peers to one receiver is not described in detail. In simulation, three selections strategies are not compared. One quality of video is transmitted and only bandwidth resources is considered.

Rate allocation for Multi-User Video Streaming over Heterogenous Access Networks.

By X. Zhu

The paper proposed a rate allocation algorithm for for multiple-user video streaming basing on avaiable bit rate (ABR) and RTT and distortion-rate (DR) characteristics. The alg can reduce the average packet loss ratio and increase the quality video.

Basing on AIMD, they proposed two heuristics methods:
greedy and rate proportional AIMD.

latex letter vs a4 size.

http://amath.colorado.edu/documentation/LaTeX/reference/faq/a4.html

Performance Analysis of a Hybrid Media Streaming system

by Y. Tu

This paper has the following contributions:
(1)proposed an equation to describe the system throughput.
(2)accomplish a quantitative analysis of performance for single and multiple
media files access.
(3)prove the optimality of system architecture in terms of server-peer transition time and server bandwidth allocation.
(4)propose the lifespan of node in the peer failure model


Limitations:
This model is only suitable for Video on demand instead of live streaming.
The peer failure model just cover partial scenarios of real p2p streaming.

syncast

　P2P技术是一种用于不同PC用户之间、不经过中继设备直接交换数据或服务的技术，其网络通信方式如图1所示。它打破了传统的 Client/Server模式，在对等网络中，每个节点的地位都是相同的，具备客户端和服务器双重特性，可以同时作为服务使用者和服务提供者。由于 P2P技术的飞速发展，互联网的存储模式将由目前的“内容位于中心”模式转变为“内容位于边缘”模式，改变Internet现在的以大网站为中心的状态， 重返“非中心化”，将权力交还给用户。
　　P2P技术有着广阔的应用领域，目前主要的应用有文件交换、分布式计算、协同工作、分布式搜索和电子商务等。

P2P模式的变化经历了集中式、分布式和混合式3个阶段。

　　集中式P2P模式由一个中心服务器来负责记录共享信息以及反馈对这些信息的查询；每一个对等实体要对它所需共享的信息以及进行的通信负责，根据需要下 载它所需要的其他对等实体上的信息。这种形式具有中心化的特点，但是它不同于传统意义上的Client/Server模式。因为传统意义上的 Client/Server模式采用的是一种垄断的手段，所有资料都存放在服务器上客户机只能被动地从服务器上读取信息，并且客户机之间不具有交互能力； 而集中式P2P模式则是所有网上提供的资料都存放在提供该资料的客户机上，服务器上只保留索引信息，此外服务器与对等实体以及对等实体之间都具有交互能 力。
　　集中目录式P2P模型还存在很多问题，主要表现为：
　　中央服务器的瘫痪容易导致整个网络的崩溃，可靠性和安全性较低；
　　随着网络规模的扩大，中央目录服务器维护和更新的费用将急剧增加，所需成本过高；
　　缺乏有效的强制共享机制，资源可用性差。
　　与集中式P2P网络不同，分布式P2P网络不存在中枢目录服务器，或者说把所有机器都变成了服务器，这就是第二代P2P——分布式P2P。

在分布式P2P中，对等机通过与相邻对等机之间的连接遍历整个网络体系。每个对等机在功能上都是相似的，并没有专门的服务器，而对等机必须依靠它们所在的分布网络来查找文件和定位其他对等机。

分布式对等网络模型也存在很多弊端，主要表现在以下方面：

　　搜索请求要经过整个网络或者至少是一个很大的范围才能得到结果，因此，这种模式占用很多带宽，而且需要花费很长时间才能有返回结果。
　　随着网络规模的扩大，通过扩散方式定位对等点及查询信息的方法将会造成网络流量急剧增加，从而导致网络拥塞，最终使分布式P2P网络被分片，使得查询 访问只能在网络很小的范围内进行，因此，网络的可扩展性不好，不适合大型网络。
　　纯分布式的P2P模式很难被企业所利用，因为它缺少对网络上的用户节点数以及对他们提供的资源的一个总体把握。
　　安全性不高，易遭受恶意攻击，如攻击者发送垃圾查询信息，造成网络拥塞等。

混合P2P网络

　　集中式P2P有利于网络资源的快速检索，并且只要服务器能力足够强大就可以无限扩展，但是其中心化的模式容易遭到直接的攻击；分布式P2P解决了抗攻 击问题，但是又缺乏快速搜索和可扩展性。混合式P2P结合了集中式和分布式P2P的优点，在设计思想和处理能力上都得到了进一步的优化。它在分布式模式的 基础上，将用户节点按能力进行分类，使某些节点担任特殊的任务。这些节点共分为3种：
　　用户节点：普通节点，它不具有任何特殊的功能。
　搜索节点：处理搜索请求，从它们的“孩子”节点中搜索文件列表，这些节点必须有128kbit/s RAM的网络连接速度，建议使用高性能的处理器。

　　索引节点：连接速度快、内存充足的节点可以作为索引节点。索引节点用于保存可以利用的搜索节点信息，并搜集状态信息，维护网络结构信息。

　　一个节点可以既是搜索节点又是索引节点。用户节点可以选择3个搜索节点作为它的“父”节点，如果“父”节点接受该用户节点作为它的“孩子”的话，那么 该用户节点就可以提交其所要共享的列表给它的“父”节点。在缺省的情况下，搜索节点可以最多维护500个“孩子”节点。在第三代P2P的软件体系结构中， 采用了混合式P2P。这种模式的关键之一是引入了索引节点，索引节点不会直接连接到有版权的资料上，它就像搜索引擎一样，只是搜索和所需资料相关的地址， 至于用户到底连接下载了什么内容则和它无关。这种模式的关键之二是引入搜索节点，搜索节点管理着所属用户的文件列表。用户节点通过索引节点获得搜索节点信 息，之后用户节点就与获得的搜索节点相连，每一次查询都通过该搜索节点进行。当用户发出搜索请求后，如果和用户节点直接相连的搜索节点查询结果达到100 个（这里的100个搜索结果，可以由用户自己来设定）就停止；如果不足100个，就向相邻的搜索节点发出请求，如果查询结果还不够，就继续向外快速发散， 直到所有的搜索节点都被搜索到为止。若所有的搜索节点都被访问过，就意味着整个网络上的节点都被搜索到了，其速度要比纯P2P模式快得多。

　　SynaCast就是第三代混合式P2P网络的典型代表，顺应了新一代P2P的潮流，它批判地继承了前辈产品的优点，将中心目录服务器的稳定性同优化 的分布式文件管理结合起来，从而在效率上远远超出了目前已有的绝大多数P2P网络系统。应用在直播及点播领域可以完全取代传统的C/S架构的直播及点播系 统。

Scalable application-level anycast for highly dynamic groups

by M. Castro et al.
An application-level of anycast for highly dynamic groups are constructed.
In this paper, group trees are constructed by using Scribe.
It enables management and location of dynamic resources in large scale p2p system.

Considering resources, CPU, bandwidth and storage are discussed.
(1)the nodes which have a job form a queue when the CPU demand is larger than CPU supply; otherwise, the nodes which have spare CPU form a queue.

Prive: Anonymous Location-base queries in Distributed mobile systems

by G. Ghinita

This paper proposed a decentralized architecture for preserving the anonymity of users issuing spatial queries to LBS. Prive distributes its state among all the peers.

Prive use k-anonimous technology to conceal the spatial-location of mobile device.

ffmpeg

How to use ffmpeg to conduct a video conversion?
An example:
ffmpeg -benchmark -s 352x288 -i foreman.yuv -g 110 -bf 6 foreman.mpg

-g group number
foreman.yuv can be downloaded from arizona university website.

Brado: scalable streaming through reconfigurable trees

by J. He
Brado address the scalability (overhead scalability and performance scalability) challenges posed by system dynamics. This paper model the effect of content switching and link dynamics on TCP window evolution and buffers.
Three-fold contributions
(1)architecturely: proposed the branching point concept to maintain the same out-degree.
(2)Analytically, lower bound throughput is independent of group size
(3)Practically, Brado is simulated and is showed the throughtput stops decreasing after a certain group size and delay grows log.ly with subscribing nodes.

Limitations:
need to be added.

Using Randomized Techniques to Build Scalable Intrusion-Tolerant Overlay Networks.

By R. Renesse
In this paper, Fireflies, a scalable protocol for supporting intrusion-tolerant overlay network is proposed.

First, this paper pointed that DHT is not secure and efficient although it is popular.
Firefly has a complete view of its live peers and it has four randomization techniques.
(1) a broadcast channel based on gossip
(2)a limit membership views for gossip
(3)a membership protocol
(4)an adaptive failure detection protocol

In conclusion, fireflies can tolerate Byzantine memebers and used as an intrusion-tolerant DHT supporting data sharing??

limitation:
The detail of fireflies protocol is not discussed.
Overhead of communication and verification need to be considered. How to decrease that overhead?

A survey of application layer techniques

http://www.cse.wustl.edu/~jain/papers/ftp/amjrts99/index.html

techniques for efficient streaming of Layered video in Heterogeneous client environments

by Raghuveer A. et al.

This paper combine layered coding and transcoding together in order to support universal multimedia acccess (UMA). The purpose for that combination is to use them in complementary roles.
for example, layered coding is not flexible enough and some extra bandwidth is wasted due to the coding and packetization overhead. In addition, no real layered coding system is deploied. Transcoding is computing intensive.

A new architecture that could build on the strengths of both layered coding and transcoding. eable efficient transport of layered video in three ways: (1) use transcoding. (2)Use transcoding to complement layered coding. (3) reduce layering overhead.

Limitations:
(1)The idea is new. but is it necessary to combine layered coding with transcoding? For VoD server, storage is not a big problem.
(2)The transcoding cost from version x to version y seems simple.

mTreebone: A hybrid Tree/Mesh Overlay for Application-Layer Live video multicast

In this paper, the authors proposed a novel hybrid tree-mesh design that leverages both overlays. The key ideas is to identify a set of stable nodes to construct a tree-based backbone, called tree bone. These stables trees form a mesh overlay which can accommodate node dynamics and exploit the available bandwidth between overlay nodes.

Tree optimization
(1)High degree preemption
(2)low delay jump

Limitations:
(1) how to find the stable nodes? In optimal stable node identification. the stability of a node is proportional to its duration in the overlay. The longer the node stays in the system, the node is more stable?
(2) The overlays setup protocol is not discussed in detail.
(3)how about there is not enough stable nodes to support a large number of out-skirt nodes?