CS 269 I : Incentives in Computer

نویسنده

  • Tim Roughgarden
چکیده

Ten years ago, the design of peer-to-peer networks was one of the hottest topics in systems research. Less so these days, but peer-to-peer networks remain practically relevant and also provide an intriguing case study of how a system evolves in response to incentive issues. The story begins in 1999, with Napster. The primary purpose of Napster was to provide a centralized and searchable directory listing which users have copies of various files (e.g., mp3s). Prior to this, Internet users exchanged files in a more ad hoc fashion, for example via newsgroups. Napster’s role was that of a match-maker: it matched up people who wanted a file with people who already have it. The file transfer itself was then done directly between the matched users. Note that a user generally acts as a server for some file transfers, and as a client for others (in contrast to the traditional client-server model). In 2000, lawsuits against Napster started rolling in (for copyright infringement). Most famously by the RIAA, but also directly by artists, including Metallica (pissing off all their fans) and Dr. Dre. After Napster failed to comply with requests to block access to copyrighted material, they were shut down in 2001. Napster’s spectacular rise (up to around 25M users) clearly demonstrated the demand for peer-to-peer networks. Its shutdown by the government highlighted the vulnerabilities of a centralized system (like Napster’s directory), so the next step was to develop completely decentralized peer-to-peer networks. In the new systems, each user maintains connections to a group of peers, and searching for a file now involves searching through the network of connections (by breadth-first search, in effect). Once a user with the desired file is located, the file transfer is done directly between the two users. There were several such systems; a major one was Gnutella, which debuted in 2000. The fully decentralized nature of this ∗ c ©2016, Tim Roughgarden. †Department of Computer Science, Stanford University, 474 Gates Building, 353 Serra Mall, Stanford, CA 94305. Email: [email protected]. Others at that time included the now-forgotten FastTrack and eDonkey. Many clever follow-up designs followed throughout the mid-aughts.

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

CS 269 I : Incentives in Computer Science Lecture # 6 : Incentivizing Participation ∗

Last lecture, we discussed the broken incentives in Gnutella, where a Prisoner’s Dilemmatype scenario gives users a strong incentive to free ride. This is one version of a broader issue, that of incentivizing participation. (In Gnutella, we can equate uploading with participating.) As you can imagine, incentivizing high levels of participation is a first-order concern for many systems and busin...

متن کامل

CS 269 I : Incentives in Computer Science Lecture # 15 : The VCG Mechanism ∗

First we prove the important result stated last lecture about sponsored search auctions. Recall the setup: there are k slots with click-through-rates (CTRs) α1 ≥ α2 ≥ · · · ≥ αk, and each bidder i = 1, 2, . . . , n has a private valuation per-click vi. (Bids and prices are also per-click.) Our first attempt to generalize the Vickrey auction to sponsored search results in the “Generalized Second...

متن کامل

CS 269 I : Incentives in Computer Science Lecture # 16 : Revenue - Maximizing Auctions ∗

Thus far, we’ve focused on the design of auctions that maximize the social welfare (at least in the case where bids are truthful). Welfare maximization is a first-order concern in many scenarios, and as we’ve seen, this objective has guided the design of auctions for sponsored search and online advertising more generally. Revenue is generated in welfare-maximizing auctions, but only as a side e...

متن کامل

CS 269 I : Incentives in Computer Science

1. Rank aggregation. Here, voters correspond to different ranking algorithms. Consider a prediction problem where the goal is to come up with a ranked list, such as the Web pages most relevant to a search query. Imagine you have several different heuristics for producing such ranked lists. Maybe one heuristic is based on anchor text, another on link structure, and another on the page content. I...

متن کامل

CS 269 I : Incentives in Computer Science Lecture # 2 : Stable Matching ∗

Last lecture, we turned a critical eye to the way college admissions are run in the U.S. Could the system be improved? For example, could there be a sensible centralized clearinghouse, like a two-sided analog of the Draw? If so, what would it look like? For inspiration, we are looking at the National Resident Matching Program (NRMP), who had to solve a similar problem to assign newly minted med...

متن کامل

Do Gender Differences in Perceived Prototypical Computer Scientists and Engineers Contribute to Gender Gaps in Computer Science and Engineering?

Women are vastly underrepresented in the fields of computer science and engineering (CS&E). We examined whether women might view the intellectual characteristics of prototypical individuals in CS&E in more stereotype-consistent ways than men might and, consequently, show less interest in CS&E. We asked 269 U.S. college students (187, 69.5% women) to describe the prototypical computer scientist ...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

عنوان ژورنال:

دوره   شماره 

صفحات  -

تاریخ انتشار 2016