Communications of the ACM
Home/Magazine Archive/February 2003 (Vol. 46, No. 2)/An End-User Perspective on File-Sharing Systems/Full Text
Technical and social components of peer-to-peer computing
An End-User Perspective on File-Sharing Systems
Because resources are shared without central management, P2P groups can form and change spontaneously. Such a group can easily extend beyond a fixed organizational boundary. Together these features help reduce the gaps (such as individual versus group work or local- versus wide-area network), that need to be filled for successful group work [2] and thus facilitate sharing resources among knowledge workers, such as knowledge of the workers closest to the tasks at hand previously more difficult to share.
This article examines P2P file-sharing systems from an end-user perspective. To date, these systems were studied mostly from a technical or legal perspective. Yet understanding what aspects of these systems really matter to the user is critical for designing better systems. In the last two years, about half a billion dollars have been invested in the companies that build P2P systems [9]. Many of these systems build on the features of file-sharing systems. For success, these companies need to do more than solve technical challenges. They need to know what their potential users consider important.
An overall picture of how current users perceive existing file-sharing systems, the importance of various features, and the concerns underlying these reactions.
For example, take the problem of "free riders." A network analysis shows that 70% of system users only download files without uploading any [1]. These free riders, or freeloaders as they are called in this context, have been labeled "Net leeches" by some and viewed as a potential cause for the downfall of P2P systems. If so, should systems be equipped with an anti-freeloading mechanism such as tying the number of downloads to the number of uploads or using a point system that rewards contributors? Is it possible that most users do not consider freeloading to be a problem? Or is it that an anti-freeloading mechanism might introduce psychological effects that will actually work against the success of these systems [7]?
As another example, new fee-based applications like MusicNet are expected to charge about $10 per month with a fixed number of allowable downloads while reducing the download time to 10 seconds for a typical file, down from four or five minutes with current no-fee systems. Understanding if enough end users exist for whom this tradeoff is reasonable is critical for the success of such applications. Examination of P2P-related Usenet groups reveal strong opinions about the importance of being able to share files without fees. Do these opinions stem from ideology (the importance of preserving freedom to exchange) or pragmatism? Is there a substantial enough percentage of the user base willing to pay to make this emerging business model successful? How important are community features such as chat or buddy lists that many of these systems provide? And do users have a sense of community?
This article presents the results of a study designed to give an overall picture of how current users perceive existing P2P file-sharing systems, how important different features are these users, and what concerns underlie these perceptions. The findings are built upon the results from a survey, qualitative analysis of P2P Usenet messages, and individual interviews.
The Study
The details of the study, including its procedure and the findings, are reported elsewhere [4]. Here they are only described selectively. Fifty-nine systems in this category were identified using magazine and Web resources (see [7]). The features of each system were then identified using its published reviews and merged into a master list. Two coders then independently categorized each feature in this master list into more general categories and the results are compared. This process was repeated until a consensus was met on the general categories of the features. In order to test the completeness of this list, the two coders independently examined all the messages from the two main Usenet groups on P2P systems from February 2000 to September 2001 and identified all the system features mentioned in these messages. They then categorized these features into one or another of the features on the list. This process identified two additional features. The final 26 categories are shown in the figure. A survey was designed with each of these features as an item the respondent was asked to rate by importance on a scale of 17. Administered to students at the University of Colorado and industry practitioners, the final analysis used 242 complete responses. To identify the factors underlying user perception, a principal component analysis was run on the result of the survey. The textual analysis of the P2P Usenet group messages and interviews with individual users also provided qualitative data on user perception.
Results
The table lists the features by the order of importance as revealed by the survey. The figure shows the eight factors identified by the principal component analysis as underlying the 26 features. Each of these factors is shown as a box containing the individual features that belong to the factor. These factors and the associated features are discussed from left to right and then top to bottom.
As shown in the table, by far the most important feature for the end user is that no fee is charged for downloading the files. One source of this feature's perceived importance seems ideological as indicated by strong comments found on the survey and the Usenet.
Another source of this perceived importance for the no-fee feature seems pragmatic. The principal component analysis factors this feature with three other features: Has large file selection, Has large user base, and Supports only legal files. One interpretation of this factoring is that these four features together capture the core product value of the file-sharing systemnamely being able to get as many desired files as possible. Under this interpretation, Has large file selection and Has a large user base would serve the goal of getting as many desired files as possible by increasing the available selections, and Charges no fee would serve the goal by freeing the user from any financial constraint. Supports legal files would undermine this goal by restricting the size of available selections, as is the case in the analysis.
The second factor we discuss contains the features ranked second through fourth in importance, which all reflect concerns about the basic properties of the process infrastructure. The high ranking in importance of these features indicates the critical nature of the basic infrastructure in P2P systems as in any other system. The next three factors contain six features that capture the enhanced value of the product. For example, features of the product that are absolutely essential but make its use easier. Can exit nicely and Can organize files as library are grouped into a factor. Has good search feature and Provides server information are factored into another. Can resume loading and Has good security feature are factored into the third. The three factors rank closely together on average. Of the features, Can resume downloading ranks highest, reflecting the major part downloading plays in the overall process, the frequency with which download processes are interrupted, and perhaps the frustration one can experience when this occurs.
The next three factors address different aspects of the social concerns. The factor containing Can control spam, Can turn off ad, and Can filter content reflect the importance of not being interrupted by unwanted materials or privacy concerns. Another factor, consisting of Supports direct messaging, Supports chat, Supports buddy list, Has voice connection, Supports passive search and Has colorful interface, reflects the importance of community because the features, except for the last, represent different ways in which users can communicate or are grouped with one another. The low rank of the features in this factor is somewhat surprising given the roles they can play in learning, debugging, sharing information, and/or fostering community spirit. A significant difference exists between more experienced users and less experienced users concerning the importance of this factor, suggesting users find these features useful when learning to use the system, but its importance diminishes as they become more comfortable with the system.
If everyone on campus turned off the outbound KaZaA traffic, approximately 50% more bandwidth could be freed for other Internet traffic.
Another group includes Can credit contributors, Has points for uploading, and Is open source. This group is labeled the "equity factor" because the first two features reflect the importance users feel about rewarding those who contribute files; the topic of the third, open source, has been highly associated with the equity issue. Given the publicity the equity and reward issues receive from media, whether for the original artists, distributors, or those who upload, it is also surprising that these features in this factor rank the lowest. However, those survey respondents who upload the most often value these features at a higher rate than those who rarely upload. So it does not seem that those who upload are purely altruistic if altruism means being satisfied simply to see their files downloaded by others.
The figure provides an overall picture of how important the feature categories are and how they relate to traditional software requirements [5]. The overall goal of getting desired files with maximum satisfaction is shown at the top of the figure. Individual features are shown in rounded rectangles. Features whose importance average was highest (>6) are shown with a red boundary, the next highest (>5) in green, the next (>4) in blue, and the next (>3) in gray. Requirement categories such as "download desired file," "upload desired file," and "minimize cost" are shown in rectangles. Arrows link a feature or a requirement to another in a "serve" relation [3]. For example, "download desired files" > "share desired files with maximum satisfaction" or "has large file selection" > "download desired files" means the former feature serves to satisfy the latter requirement. When a feature or a requirement has multiple serve relations, only major ones are shown to reduce cluttering of the diagram. The eight factors revealed by the principal component analysis are shown by the eight boxes, each of which surround the features that each factor contains. The limited space of this article prevents the presentation of the rationales underlying this picture. (They are reported in [4], which also includes other findings of the study such as gender and group differences, such as novice vs. experienced, high vs. low self-efficacy or self-confidence with regard to technologies.)
Interpretations
How does this composite of the user help us answer the questions raised regarding freeloading and the fee-based business model?
Freeloading does not seem to be a major issue for most users including those who upload. The features Can credit contributors and Has points for uploading rank the lowest in importance, even for those who upload, although uploaders believe so more than noncontributors. It is as if they are saying: "That would be nice but there are more important things such as no fee, reliability, and file selection."
Why, then, do people upload files? Probably not out of sheer altruism since they rank community features as low as noncontributors. The study paints a picture of the pragmatic user. They might upload files because they fear the system will collapse unless they do. They refuse to risk such a collapse by waiting for others to supply them while they themselves take a free ride. They'd rather take the small trouble of creating and uploading files, trusting that enough others will do so also. As long as these users get what they need, it will not cost other people who free ride because shared files are replicable with little cost other than resources already paid for, such as disk space, computing power, and Internet access. This is called the "Cornucopia of Digital Commons" [6], and is the real genius, more than any technical breakthrough, behind the success of P2P systems.
If we accept this interpretation, then a large user base is critical for two reasons. First, it increases the number of contributors and, hence, the size of file selections. Secondly, it increases the pool of resources; even though free riders do not contribute files, their computers act as servers and routers in sharing these files. In this light, FastTrack, the once- Amsterdam-based company that has been bought by the Australian-based Sharman Networks, made a smart move in combining the user base of three different popular applications using its network technology. According to www.slyck.com (June 2002), FastTrack is said to have an online user base of over 1.7 million, representing a dramatic rise in the user base.
If so, an anti-freeloading mechanism would deter instead of promote the success of the system and limit the user base. Such a mechanism can cause user frustration when, for example, download privileges are tied to uploaded contributions. It might prompt users to upload junk files to get around such mechanisms. Or such a regulatory measure could result in a psychological transaction cost [7], such as when a user has to assign desirability points for a given file. It might also induce the perception of inequity even among the contributors if it makes them feel they deserve more recognition than what they receive. Whatever incentives such a mechanism is designed to provide may not warrant these potential costs, especially when the importance of recognition ranks so low even among the contributors.
There are other intriguing issues that deserve our attention. According to the analysis of the network traffic at Cornell University during February 2002 www.cit.cornell.edu/computer/students/ bandwidth/charts.html), over half the used bandwidth is from KaZaA, the most popular P2P application currently used. Of this half, about two-thirds of the traffic is outbound, implying that if everyone on campus turned off the outbound KaZaA traffic, approximately 50% more bandwidth could be freed for other Internet traffic. Furthermore, it points out that the single heaviest user used the same amount of bandwidth as 2,800 typical users. Further studies are in need in order to illuminate whether the awareness of these facts influences the user behavior.
What does this tell us about the future of the new fee-based business model of file-sharing applications? As mentioned earlier, this fee can fund better performance features for the new model. Given the high ranking of these features, users might buy into it provided the fee is reasonable. However, there is a potential showstopper for this business model. Because the record labels and publishing companies do not own the rights to music from the most popular artists, it might be difficult to provide a library nearly as complete as the current free systems. Given the importance of the large file selection and its higher importance for experienced users, ensuring a large available selection seems critical for success.
With the recent attention and promises of P2P systems, many more will be built in the future. These systems, ranging from collaboration tools, distributed search engines, gaming, reputation management, and writable webs, will be based on many of the file-sharing system features. The designers of these systems, whether motivated by profit or service, should be aware of what matters to their potential users.
This article has reported a study that sheds some light on the user perception of P2P systems. Although not discussed here, the study also reveals group differences (social features that females and novices perceive to be of higher importance) and the way in which users appropriate the system's features (adapted in a way not necessarily intended by the designer). It also points to the importance and the need for further studies like this one and the ones pursued by human computer interaction researchers that examine end-user perceptions of technologies, especially innovative technologies such as P2P systems.
References
1. Adar, E. and Huberman, B.A. Free riding on Gnutella. First Monday 5, 10, 1998
2. Baecker, R., Ed. Readings in Groupware and Computer-Supported Cooperative Work. Morgan Kaufmann, San Francisco, CA, 1993.
3. Fischer, G., et al. Making argumentation server design, in Design Rationale: Concepts, Techniques, and Use. T.P. Moran and J.M. Carroll, Eds. Lawrence Erlbaum Associates, Mahwah, N.J, 1996.
4. Lee, J. Peer-to-Peer File Sharing Systems: What Matters to the End. MIT Center for Coordination Science, Rep. No. 217, Cambridge, MA.
5. Maciaszek, L.A. Requirements Analysis and System Design: Developing Information Systems with UML. Addison-Wesley, Reading, MA. 2001.
6. Oram, A., Ed. Peer-to-Peer: Harnessing the Power of Disruptive Technologies. O'Reilly & Associates, Cambridge, MA, 2001.
7. O'Reilly & Associates (Nov. 14, 2001); www.openp2p.com.
8. Schonfeld, E. Future boy: The Napster legacy. Business, 2.0. (Sept. 25, 2001).
9. Shirky, C., Truelove, K., Dornfest, R., and Gonze, L. P2P Networking Overview. The Emergent P2P Platform of Presence, Identity, and Edge Resources. O'Reilly & Associates, 2001.
Figures
Figure. The eight groups of the P2P file-sharing system features and their relationship to the traditional software requirement categories.
Tables
Table. Representative features of P2P file-sharing systems in the order of perceived importance.
©2003 ACM 0002-0782/03/0200 $5.00
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee.
The Digital Library is published by the Association for Computing Machinery. Copyright © 2003 ACM, Inc.
No entries found