Pipelined Algorithms to Detect Cheating in Long-Term Grid Computations

📝 Abstract

This paper studies pipelined algorithms for protecting distributed grid computations from cheating participants, who wish to be rewarded for tasks they receive but don’t perform. We present improved cheater detection algorithms that utilize natural delays that exist in long-term grid computations. In particular, we partition the sequence of grid tasks into two interleaved sequences of task rounds, and we show how to use those rounds to devise the first general-purpose scheme that can catch all cheaters, even when cheaters collude. The main idea of this algorithm might at first seem counter-intuitive–we have the participants check each other’s work. A naive implementation of this approach would, of course, be susceptible to collusion attacks, but we show that by, adapting efficient solutions to the parallel processor diagnosis problem, we can tolerate collusions of lazy cheaters, even if the number of such cheaters is a fraction of the total number of participants. We also include a simple economic analysis of cheaters in grid computations and a parameterization of the main deterrent that can be used against them–the probability of being caught.

💡 Analysis

This paper studies pipelined algorithms for protecting distributed grid computations from cheating participants, who wish to be rewarded for tasks they receive but don’t perform. We present improved cheater detection algorithms that utilize natural delays that exist in long-term grid computations. In particular, we partition the sequence of grid tasks into two interleaved sequences of task rounds, and we show how to use those rounds to devise the first general-purpose scheme that can catch all cheaters, even when cheaters collude. The main idea of this algorithm might at first seem counter-intuitive–we have the participants check each other’s work. A naive implementation of this approach would, of course, be susceptible to collusion attacks, but we show that by, adapting efficient solutions to the parallel processor diagnosis problem, we can tolerate collusions of lazy cheaters, even if the number of such cheaters is a fraction of the total number of participants. We also include a simple economic analysis of cheaters in grid computations and a parameterization of the main deterrent that can be used against them–the probability of being caught.

📄 Content

Pipelined Algorithms to Detect Cheating in Long-Term Grid Computations Michael T. Goodrich Dept. of Computer Science, Univ. of California, Irvine, CA 92697-3235 USA Abstract This paper studies pipelined algorithms for protecting distributed grid computations from cheating participants, who wish to be rewarded for tasks they receive but don’t perform. We present improved cheater detection algorithms that utilize natural delays that exist in long-term grid computations. In particular, we partition the sequence of grid tasks into two interleaved sequences of task rounds, and we show how to use those rounds to devise the first general- purpose scheme that can catch all cheaters, even when cheaters collude. The main idea of this algorithm might at first seem counter-intuitive—we have the participants check each other’s work. A naive implementation of this approach would, of course, be susceptible to collusion attacks, but we show that by, adapting efficient solutions to the parallel processor diagnosis problem, we can tolerate collusions of lazy cheaters, even if the number of such cheaters is a fraction of the total number of participants. We also include a simple economic analysis of cheaters in grid computations and a parameterization of the main deterrent that can be used against them—the probability of being caught. Keywords: Grid computing, pipelined algorithms, parallel algorithms, security. 1 Introduction One of the success stories of parallel and distributed algorithms is the computational grid paradigm for solving large computational problems. In this paradigm, a supervisor distributes a set of inde- pendent tasks to a community of participants, who perform those tasks and send back the results. (See Figure 1.) Examples of well-known on-going grid computations include SETI@home, which claims over 7 million participants who have collectively performed over 1.5 billion tasks aimed at finding intelligent patterns in extraterrestrial signals, and Grid.org, which claims over 3 million computers being used to solve large scientific problems related to medicine. The participants in grid computing environments are typically volunteers rewarded with recog- nitions of their service. For example, SETI@home regularly posts the names of its top 1000 users. Unfortunately, even with such modest rewards, grid computations must deal with cheaters. In- deed, the director for SETI@home is quoted [15] as saying that their project spends half of their resources dealing with cheaters, who comprise roughly 1% of their users. He mentioned that some 1 arXiv:0906.1225v1 [cs.CR] 5 Jun 2009 Figure 1: Illustrating the grid paradigm whereby a supervisor partitions a computation into tasks that are then farmed out to participants, who are expected to perform those tasks and submit re- sponses back to the supervisor. users have modified the SETI@home software to make it look like they have performed more work than they actually did. To deal with such cheaters, the SETI@home director mentioned that their system duplicates every task and sends it to two different participants for confirmation. If the two results match, they accept the computation; if the results don’t match, they send the computation to a third participant for determination of which of the original two cheated. Such duplication creates waste in the system, of course, and, even with this extra cost, it still allows for cheating if participants collude. Thus, we would ideally like to find more efficient cheater deterrents that significantly discourage cheating users even if they collude. The problem of dealing with cheaters in a grid computation becomes even more serious, of course, when the rewards for participation become more tangible. Therefore, an important challenge is the efficient detection and deterrence of colluding users who wish to be rewarded for grid tasks they don’t actually perform. 1.1 Previous Related Work The research area of designing grid protocols for detecting cheating users is known as uncheatable grid computing [10, 11, 13, 14, 22]. Roughly speaking, previous approaches have relied either on the general-purpose approach of replicating tasks, which creates multiplicative overheads (as in the SETI@home approach), or on special-purpose ad hoc solutions. Golle and Mironov propose a special-purpose ringer scheme [13], which is restricted to the inversion of one-way functions. Szada, Lawson, and Owen extend their scheme somewhat [22], but their solutions still are not gen- eral purpose. Du and Goodrich [10] show how to utilize chaff-injection techniques to reduce the ability of lazy cheaters to collude on outlier-search computations. Likewise, Du et al. [11] propose 2 ad hoc checking algorithms for cheater detection in grid computations. In their approach, the su- pervisor randomly selects and verifies for himself some samples from the task domain D assigned to a participant. This approach places a significant computational burden on the supervisor in ad- dition

This content is AI-processed based on ArXiv data.