The Digital Economy: How Digital Goods are Reshaping the Rules of Commerce
Emergence of Peer-to-Peer

By Eric de Fontenay (Founder & Publisher) (more articles from this author) 2000-11-01

Comment | Email | Print | RSS

Much has been said about the phenomenal growth of processing power, often citing Moore's Law. This is also a characteristic shared with growth in bandwidth, but equally points to its weakness: that even these growth rates barely satisfy the growing requirements of existing and emerging consumer and business applications. But if one considers the increasing demand for high performance computing requirements, they are on an entirely different scale.

A first step in addressing the problem has been computers offering a multiprocessor platform on a single machine. This has been a strategy actively pursued by Apple with its G4 PowerPC chip in order to retain its role in the multimedia and content production sector. But this demanded special "glue" chips or one-of-a-kind interconnection schemes. Massively parallel processing (MPP) brings the idea of multiprocessor platform machines to the next extreme by fitting up to up to 2048 processors on a single machine, typically referred to as a supercomputer.

To understand the relationship between MMP and peer-to-peer (P2P) networking, it is useful to consider the accomplishment of the Beowolf Project. Beowolf was the actual first implementation of using commodity off the shelf (COTS) base systems to satisfy specific computational requirements, where 16 DX4 processor machines were clustered into a high performance network of standard PCs under the Earth and Space Sciences (ESS) project for NASA. The cluster allowed for greater than a gigaflop/s sustained performance at a fraction (less than $50,000) of the cost of a supercomputer. The success of the Beowulf project that has since extended beyond NASA demonstrated the power of networking as an alternative to the expense and complexity of dedicated supercomputing power.

Perhaps the beat known and most relevant implementation of the Beowolf concept is the SETI@home project. The SETI Project (Search for ExtraTerrestrial Intelligence) is dedicated to analyzing all incoming radio signals from space for patterns that might provide an indication of intelligent life. But it generally faced a problem of a lack of resources to meet the processing power required to analyze all the incoming radio transmissions. To address the problem, a 300 Kb screensaver incorporating a software analyzing chunks of spectrum from the Arecibo Observatory in Puerto Rico while the client machine was idle was made available over the Internet. The resulting analysis from each client machine would be sent to SERENDIP (Search for Extraterrestrial Radio Emissions from Nearby Developed Intelligent Populations), supplementing the efforts of the spectrum analyzers on location. This has allowed the SETI@home project to provide an additional 13.5 trillion floating-point operations per second from individuals downloading the special screen saver¹.

A notable difference between both projects that should be mentioned. In the case of Beowulf, the nodes in the cluster are dedicated to the cluster while the nodes in the SETI@home project is 'casually' dedicated to the SERENDIP lab at Berkeley University. By only dedicating nodes on the SETI@home network while the nodes are idle, the project has turned what would are client machines into part-time servers. This allows for more efficient use of otherwise underutilized computing power at little cost to all the nodes on the system, dispensing with the need for dedicated resources. This allows the project to capitalize on the networked computer aspect of MPP & COTS, using it to achieve far grater computational power than is possible in the Beowulf project, despite the 'casually dedicated' nature of the model.

P2P though extends well beyond distributing computational power as demonstrated by software like Napster. Napster is a software program that allows its users to 'share' mp3 files with other members of the network. The Napster software records and files the playlist of mp3 files to be shared with the Napster community on a centralized set of servers. As users search the directory of file names to locate and request the desired file, the software makes a one-to-one connection between the two computers, allowing for the transfer of the file. It is important to note though that Napster sole purpose is as a distribution point for the software and storing/maintenance of the directory. Its does not copy or host any files itself. As such, it simply acts as a mediator between the various nodes on the network between the various nodes on the network.

The implications of the emergence of digital goods and new distribution models such as P2P are simply revolutionary. As illustrated by earlier, the Internet represents a significant centralization from multiple inventory and retail points in the physical world to the single server-multiple client model. This leads to certain weaknesses with the model including bandwidth management issues. P2P though, by its very decentralized nature, alleviates the problem by spreading demand throughout the network. Thus, although specific nodes may be overburdened, the rest of the network will remain accessible. It should be emphasized that Napster is not a centralized P2P system in that the file directory system that allows users to locate and request files is centrally located on Napster's servers. This has not only made them subject to similar load balancing issues as traditional online distribution as well as leaving the legal entity that owns the servers subject to copyright liability.

Gnutella addresses this vulnerability issue by adopting a purely decentralized system. Instead of searching a centralized set of servers as with Napster, a query from a Gnutella user searches the Gnutella Network "horizon," or virtual vicinity. The system searches a segment of the network for the desired work, typically up to 1,000 host machines. Over the period of a connection, hosts fall off and join the node's horizon, allowing the search to be effectuated over a larger group of host than the initial segment. As users transfer files to its own machine, they then becomes available to a new set of host as the machine shifts from segment to segment. There is no centralized server through which all traffic is routed. The decentralized P2P concept has been further improved by Freenet which uses intelligent routing and caching to further alleviate bandwidth load.

Footnotes

¹ "As of February, 2000, SETI@home has grown to encompass 1.6 million participants in 224 countries. The amount of computing time contributed since May, 1999 is equal to 165,000 years, averaging 10 Teraflops (about 10 times more than the largest supercomputer on the planet). It is the largest computation ever done, and has attracted the participation of 20,000 groups such as schools and private companies." http://www.seti-inst.edu/science/setiathome.html

Home » Insider Scoop » The Digital Economy: How Digital Goods are Reshaping the Rules of Commerce Permalink: http://www.musicdish.com/mag/?id=2158 Email | Print | Comment | RSS

back | top