1、外文文献翻译 原文： Peer-to-Peer Streaming Media Delivery Peer-to-Peer Architecture Whatever definitions have been put upon it, peer-to-peer is an effective rallying cry for a new way of doing things. Many do not consider it a new way； it has been argued that the current interest in peer-to-peer is merely th
2、e pendulum swing between centralized and decentralized systems. That cycle can be described as : 1. Decentralize to remove bottlenecks 2. Centralize to increase efficiency Nonetheless, the particular inflection point of resources available on the Internet at this time has allowed peer-to-peer system
3、s to exhibit remarkable scalability and resource exchange. This paper briefly describes a different way of looking at the resources available in these systems； it then illustrates the applicability of peer-to-peer systems to content delivery. Peer-to-peer is used broadly to describe a variety of net
4、work systems that generally run at the presentation, session, and application layers, although ad-hoc networks and other systems use the same concepts from the physical layer up. Specifically, peer-to-peer spans content delivery, collaboration, caching, business process automation, supply chain mana
5、gement, grid computing, distributed computation, business-to-business exchanges, data deployment, user to user communication, user communities, ad-hoc networks, and the Internet itself. Perhaps peer-to-peer should be considered more of an architectural approach than a specific technology or business
6、 approach. An way of seeing if a problem is susceptible to a peer-to-peer approach is to ask, “If every client in this system could also provide the service they consume, would there be a benefit?” It is not always the case that there is a benefit； many database applications require centralization f
7、or security and simplicity of administration, for example. Unusual Peer-to-Peer Examples The oft-cited ICQ and Napster are two pioneering peer-to-peer examples. Both provide an alternative system to DNS for naming, (an attribute of some peer-to-peer definitions ) and both provide the ability for use
8、rs to directly communicate, providing the “person-to-person” aspect also associated with peer-to-peer. There are, however, a variety of other systems providing earlier examples of the shift towards peer-to-peer systems. If peer-to-peer is considered as a quality with a gradient scale, ranging from c
9、lient-server to a more equilateral power of computing systems, any system that provides a higher ratio of servers to clients could be regarded as peer-to-peer. Quake was (and remains) a 3d online multiplayer video game. Quake (and later QuakeWorld) provided a client-server system for synchronized vi
10、deo gaming. The servers were usually high bandwidth, high powered systems, but due to the demands of online video gaming, so were the clients. And in Quake, the server was actually embedded in the client application, blurring the distinction between clients and servers and allowing any node to act (
11、by user selection) as a client or a server.Among millions of online Quake players, there eventually existed tens of thousands of servers on the Internet, and so the ratio of clients to servers began to even out. ShoutCast was designed as a plug-in to a popular MP3 player that enabled live streaming
12、of MP3 audio over an HTTP connection. Users could use ShoutCast to create a radio station (on the Internet) based on their MP3 collection. This allowed serving from a client application, and increased the ease for an end user in configuring this server. Peer-to-peer Resources Storage CPU Bandwidth S
13、torage and CPU cycles tend to be the two resources most commonly cited in peer-to-peer systems. There are, however, several resources that, isolated, can be used to better describe the range of optimizations available through peer-to-peer approaches. Bandwidth is a resource that is transient and non
14、-recapturable. In the same way that unused airline seats cannot be recaptured as a resource, bandwidth, when available and unused, is lost to time. The first stage of consumer Internet expansion involved a great disparity between client bandwidth and server bandwidth, but as broadband equalizes this
15、 resource, it becomes available to a peer-to-peer system. Presence Latency/Proximity Presence can be viewed as a resource. When the p in p2p stands for “person” (as in instant messaging scenarios), presence is the resource of that person being online and available for communication at that time. Thi
16、s enables online collaboration because it provides at a glance notification of availability. Latency and proximity are two relatively unremarked resources that are key to interactive simulations on the Internet. Quake provides an example of a large number of servers (Quake servers) located with an a
17、lternative namespace system (GameSpy) and then sorted by latency. The tens of thousands of Quake servers provided a pool of the service called “Quake”, but that service was ineffective if the number of hops between the server and the clients, or the latency, or the packet loss, was too high. It was
18、not the storage space of the machines that was being utilized (although it was some of the CPU cycles), but it was the proximity of the server to the client that was the precious resource. Napster used only megabytes of client-server traffic to manage, direct, and control terabytes of peer-to-peer t
19、raffic. The storage of these peers was notable, but the real feat was offloading the bandwidth requirements to the peers consuming the resources and coordinating their relatively seamless interchange. A centralized Napster would have been technically trivial to implement, but prohibitively costly an
20、d remarkably difficult to scale. A consistent theme in peer-to-peer systems to date is that they put additional code at the client level and thus where it can do different things than if it was centralized. A strength of peer-to-peer systems is that they distribute code that can provide services at
21、a more strategic location. For instance, some peer-to-peer systems route traffic between peers. These servers provide CPU cycles (to perform the service), proximity (if the routing algorithm is based on low hop count for instance), and bandwidth (by providing the routing). This combination of variou
22、s resources shows why web services now tend to be included in discussions of peer-to-peer architectures. Content Delivery Costs The primary resource contention on the Internet is over bandwidth. The costs of bandwidth, especially bandwidth with the quality of service goals necessary to support onlin
23、e audio and video, does not drop as dramatically as the cost of computer hardware. Another interesting aspect of bandwidth is that consumers tend to pay a flat fee or a low fee for a moderate amount of broadband bandwidth, whereas enterprise tends to pay larger, variable costs for their bandwidth. T
24、hese two factors present an opportunity for systems that can substitute low cost hardware into higher value bandwidth, or can substitute fixed-cost consumer bandwidth for variable cost enterprise bandwidth. Enough large, early streaming companies have failed because of the overwhelming cost of strea
25、ming bandwidth. Would-be Internet “television stations” were technically feasible but completely impractical from a cost standpoint. The supply chain of Internet video in particular is quite broken: Content providers are slow to advertise their services because they cannot afford the bandwidth costs
26、 of an increased audience. Similarly, bandwidth providers cater to customers who tend not to use their bandwidth. This situation tends to createunprofitable, shrinking content providers who pay too much for bandwidth they do not use. This bandwidth need applies to both static (web page) and dynamic (streaming media) content； shopping for bandwidth and constraining he costs can be difficult and can result in highly variable quality of service on the part of content providers. This is an opportunity for peer-to-peer technology.
