Agustín Santos

Reliably Executing Tasks in the Presence of Untrusted Entities

In this work we consider a distributed system formed by a master processor and a collection of n processors (workers) that can execute tasks; worker processors are untrusted and might act maliciously. The master assigns tasks to workers to be executed. Each task returns a binary value, and we want the master to accept only correct values with high probability. Furthermore, we assume that the service provided by the workers is not free; for each task that a worker is assigned, the master is charged with a work-unit. Therefore, considering a single task assigned to several workers, our goal is to have the master computer to accept the correct value of the task with high probability, with the smallest possible amount of work (number of workers the master assigns the task). We explore two ways of bounding the number of faulty processors: (a) we consider a fixed bound f < n/2 on the maximum number of workers that may fail, and (b) a probability p < 1/2 of any processor to be faulty (all processors are faulty with probability p, independently of the rest of processors). Our work demonstrates that it is possible to obtain high probability of correct acceptance with low work. In particular, by considering both mechanisms of bounding the number of malicious workers, we first show lower bounds on the minimum amount of (expected) work required, so that any algorithm accepts the correct value with probability of success 1 - epsiv, where epsiv Lt 1 (e.g., 1/n). Then we develop and analyze two algorithms, each using a different decision strategy, and show that both algorithms obtain the same probability of success 1 - epsiv, and in doing so, they require similar upper bounds on the (expected) work. Furthermore, under certain conditions, these upper bounds are asymptotically optimal with respect to our lower bounds

Adversarial Queueing Model for Continuous Network Dynamics

Abstract In this paper we initiate the generalization of the Adversarial Queueing Theory (aqt) model to capture the dynamics of continuous scenarios in which the usually assumed synchronicity of the evolution is not required anymore. We propose an asynchronous model, named continuousaqt (caqt), in which packets can have arbitrary lengths, and the network links may have different speeds (or bandwidths) and propagation delays. With respect to the standard aqt model, these new features turn out to be significant for the stability of packet scheduling policies that take them into account, but not so much for the stability of networks. From the network point of view, we show that networks with directed acyclic topologies are universally stable, i.e., stable independently of the scheduling policies and traffic patterns used in it. Interestingly enough, this even holds for traffic patterns that make links to be fully loaded. Finally, it turns out that the set of universally stable networks remains the same as in the aqt model and, therefore, the property of universal stability of networks is decidable in polynomial time. Concerning packet scheduling policies, we show that the well-known lis, sis, ftgand nfsscheduling policies remain universally stable in the caqt model. We introduce other scheduling policies that, although being universally stable in the aqt model, they are unstable under the caqt model.

RELIABLE INTERNET-BASED MASTER-WORKER COMPUTING IN THE PRESENCE OF MALICIOUS WORKERS

We consider a Master-Worker distributed system where a master processor assigns, over the Internet, tasks to a collection of n workers, which are untrusted and might act maliciously. In addition, a worker may not reply to the master, or its reply may not reach the master, due to unavailabilities or failures of the worker or the network. Each task returns a value, and the goal is for the master to accept only correct values with high probability. Furthermore, we assume that the service provided by the workers is not free; for each task that a worker is assigned, the master is charged with a work-unit. Therefore, considering a single task assigned to several workers, our objective is to have the master processor to accept the correct value of the task with high probability, with the smallest possible amount of work (number of workers the master assigns the task). We probabilistically bound the number of faulty processors by assuming a known probability p < 1/2 of any processor to be faulty. Our work demonstrates that it is possible to obtain, with provable analytical guarantees, high probability of correct acceptance with low work. In particular, we first show lower bounds on the minimum amount of (expected) work required, so that any algorithm accepts the correct value with probability of success 1 - e, where e ≪ 1 (e.g., 1/n). Then we develop and analyze two algorithms, each using a different decision strategy, and show that both algorithms obtain the same probability of success 1 - e, and in doing so, they require similar upper bounds on the (expected) work. Furthermore, under certain conditions, these upper bounds are asymptotically optimal with respect to our lower bounds.

Reliably executing tasks in the presence of malicious processors

The demand for processing large amounts of data has increased over the last decade. As traditional one-processor machines have limited computational power, distributed systems consisting of multitude of cooperating processing units are used instead. An example of such a massive distributed cooperative computation is the SETI@Home project [5]. As the search for extraterrestrial intelligence involves the analysis of gigabytes of raw data that a fixed-size collection of machines would not be able to effectively carry out, the data are distributed to millions of voluntary machines around the world. A machine acts as a server and sends data (aka tasks) to these client computers, which they process and report back the result of the task computation. This gives rise to a crucial problem: how can we prevent malicious clients from damaging the outcome of the overall computation? In this work we abstract this problem in the form of a distributed system consisting of a master fail-free processor and a collection of processors (workers) that can execute tasks; worker processors might act maliciously. Since each task returns a value, we want the master to accept only correct values with high probability. Furthermore, we assume that the service provided by the workers is not free (as opposed to the SETI@Home project). For each task that a worker executes, the master computer is charged with a work-unit. Therefore, considering a single task assigned to several workers, our goal is to have the master computer to accept the correct value of the task with high probability, with the smallest possible amount of work. We explore two ways of bounding the number of faulty processors and evaluate an algorithm that the master can run. Our preliminary analytical results show that it is possible to obtain high probability of correct acceptance with reasonable amount of work.

Graph-based Techniques for Topic Classification of Tweets in Spanish

Topic classification of texts is one of the most interesting challenges in Natural Language Processing (NLP). Topic classifiers commonly use a bag-of-words approach, in which the classifier uses (and is trained with) selected terms from the input texts. In this work we present techniques based on graph similarity to classify short texts by topic. In our classifier we build graphs from the input texts, and then use properties of these graphs to classify them. We have tested the resulting algorithm by classifying Twitter messages in Spanish among a predefined set of topics, achieving more than 70% accuracy.

Empirical Comparison of Graph-based Recommendation Engines for an Apps Ecosystem

Recommendation engines (RE) are becoming highly popular, e.g., in the area of e-commerce. A RE offers new items (products or content) to users based on their profile and historical data. The most popular algorithms used in RE are based on collaborative filtering. This technique makes recommendations based on the past behavior of other users and the similarity between users and items. In this paper we have evaluated the performance of several RE based on the properties of the networks formed by users and items. The RE use in a novel way graph theoretic concepts like edges weights or network flow. The evaluation has been conducted in a real environment (ecosystem) for recommending apps to smartphone users. The analysis of the results allows concluding that the effectiveness of a RE can be improved if the age of the data, and if a global view of the data is considered. It also shows that graph-based RE are effective, but more experiments are required for a more accurate characterization of their properties.

Performance of scheduling policies in adversarial networks with non synchronized clocks

In this paper we generalize the Continuous Adversarial Queuing Theory (CAQT) model [5] by considering the possibility that the router clocks in the network are not synchronized. Clearly, this new extension to the model only affects those scheduling policies that use some form of timing. First, if all clocks run at the same speed, maintaining constant differences, we show that all universally stable policies in CAQT that use the injection time and the remaining path to schedule packets remain universally stable. These policies include, for instance, Shortest in System (SIS) and Longest in System (LIS). Then, if clock differences can vary over time, but difference is bounded, we show the universal stability of SIS and a family of policies related to LIS. The bounds we obtain in this case depend on the maximum difference between clocks. We then present a new policy that we call Longest in Queues (LIQ), which gives priority to the packet that has been waiting the longest in edge queues. This policy is universally stable and, if clocks maintain constant differences, the bounds do not depend on them. To finish, we provide with simulation results that compare the behavior of some of these protocols in a network with stochastic injection of packets.

A Mechanism for Fair Distribution of Resources without Payments

We design a mechanism for Fair and Efficient Distribution of Resources (FEDoR) in the presence of strategic agents. We consider a multiple-instances, Bayesian setting, where in each round the preference of an agent over the set of resources is a private information. We assume that in each of r rounds n agents are competing for k non-identical indivisible goods, (n > k). In each round the strategic agents declare how much they value receiving any of the goods in the specific round. The agent declaring the highest valuation receives the good with the highest value, the agent with the second highest valuation receives the second highest valued good, etc. Hence we assume a decision function that assigns goods to agents based on their valuations. The novelty of the mechanism is that no payment scheme is required to achieve truthfulness in a setting with rational/strategic agents. The FEDoR mechanism takes advantage of the repeated nature of the framework, and through a statistical test is able to punish the misreporting agents and be fair, truthful, and socially efficient. FEDoR is fair in the sense that, in expectation over the course of the rounds, all agents will receive the same good the same amount of times. FEDoR is an eligible candidate for applications that require fair distribution of resources over time. For example, equal share of bandwidth for nodes through the same point of access. But further on, FEDoR can be applied in less trivial settings like sponsored search, where payment is necessary and can be given in the form of a flat participation fee. FEDoR can be a good candidate in a setting like that to solve the problem of starvation of publicity slots for some advertisers that have a difficult time determining their true valuations. To this extent we perform a comparison with traditional mechanisms applied to sponsored search, presenting the advantage of FEDoR.