Computing similarity between high-dimensional data is a fundamental problem in data mining and information retrieval, with numerous applications-such as e-discovery and patient similarity. To address the relevant performance and scalability challenges, approximation methods are employed. A common characteristic among all privacy-preserving approximation protocols based on sketching is that the sketching is performed locally and is based on common randomness. Inspired by the power of attacks on machine learning models, we introduce the study of adversarial inputs for secure similarity approximations. To formally capture the framework of this family of attacks we present a new threat model where a party is assumed to use the common randomness to perturb her input 1) offline, and 2) before the execution of any secure protocol, so as to steer the approximation result to a maliciously chosen output. We define perturbation attacks under this adversarial model and propose attacks for the techniques of minhash and cosine sketching. We demonstrate the simplicity and effectiveness of the attacks by measuring their success on synthetic and real data from the areas of e-discovery and patient similarity. To mitigate such perturbation attacks we propose a server-aided architecture, where an additional party, the server, assists in the secure similarity approximation by handling the common randomness as private data. We revise and introduce the necessary secure protocols so as to apply minhash and cosine sketching techniques in the server-aided architecture. Our implementation demonstrates that this new design can mitigate offline perturbation attacks without sacrificing the efficiency and scalability of the reconstruction protocol.

A lot of the large datasets analyzed today represent graphs. In many real-world applications, summarizing large graphs is beneficial (or necessary) so as to reduce a graph's size and, thus, achieve a number of benefits, including but not limited to 1) significant speed-up for graph algorithms, 2) graph storage space reduction, 3) faster network transmission, 4) improved data privacy, 5) more effective graph visualization, etc. During the summarization process, potentially useful information is removed from the graph (nodes and edges are removed or transformed). Consequently, one important problem with graph summarization is that, although it reduces the size of the input graph, it also adversely affects and reduces its utility. The key question that we pose in this paper is, can we summarize and compress a graph while ensuring that its utility or usefulness does not drop below a certain user-specified utility threshold? We explore this question and propose a novel iterative utility-driven graph summarization approach. During iterative summarization, we incrementally keep track of the utility of the graph summary. This enables a user to query a graph summary that is conditioned on a user-specified utility value. We present both exhaustive and scalable approaches for implementing our proposed solution. Our experimental results on real-world graph datasets show the effectiveness of our proposed approach. Finally, through multiple real-world applications we demonstrate the practicality of our notion of utility of the computed graph summary.

Most websites, services, and applications have come to rely on Internet services (e.g., DNS, CDN, email, WWW, etc.) offered by third parties. Although employing such services generally improves reliability and cost-effectiveness, it also creates dependencies on service providers, which may expose websites to additional risks, such as DDoS attacks or cascading failures. As cloud services are becoming more popular, an increasing percentage of the overall Internet ecosystem relies on a decreasing number of highly popular services. In our general effort to assess the security risk for a given entity, and motivated by the effects of recent service disruptions, we perform a large-scale analysis of passive and active DNS datasets including more than 2.5 trillion queries in order to discover the dependencies between websites and Internet services.

In large scale-out data architectures, data are distributed and replicated across several machines. Queries/tasks to such data architectures, are sent to a router which determines the machines containing the requested data. Ideally, to reduce the overall cost of analytics, the smallest set of machines required to satisfy the query should be returned by the router. Mathematically, this can be modeled as the set cover problem, which is NP-hard. Given large number of incoming queries in real-time, it is often impractical to compute set cover for each incoming query to perform routing. In this paper, we propose a novel technique to speedup the routing of a large number of real-time queries while minimizing the number of machines that each query touches (query span). We demonstrate that by analyzing the correlation between known queries and performing query clustering, we can reduce the set cover computation time, thereby significantly speeding up routing of unknown queries. Experiments show that our incremental set cover-based routing is 2.5 times faster and can return on average 50% fewer machines per query when compared to repeated greedy set cover and baseline routing techniques.

Lightweight virtualization, as implemented by application container solutions such as Docker, have the potential to revolutionize the way multi-tier applications are developed and deployed, especially in the cloud. The success of application containers can be partly attributed to their ability to share resources with the underlying platform that hosts them. As such, the isolation provided by such containers is not as strict as with traditional VMs. These very characteristics that have contributed to the success of application containers can also be seen as factors that limit their widespread commercial adoption, since enterprise IT administrators cannot implement the various -- and often fine-grained -- security policies they are required to abide by. This problem is of limited consequence when a host is running a single user's application containers. But sharing compute resources among multiple users is an important benefit of containers and cloud-based deployment. In this paper we present a preliminary discussion of the challenges associated with enterprise security policy management for application containers deployed in multi-user environments. Furthermore, we present Harbormaster, a system that addresses some of these challenges by enforcing policy checks on Docker container management operations and allowing administrators to implement the principle of least privilege.

Protecting data from loss is of crucial significance to businesses. Failure to protect data can lead to heavy financial and strategic losses, that are often difficult to recover from. Thus, businesses employ backup techniques to store copies of data to enable failure recovery. But surprisingly, backups often fail. Our analysis of about 48,000 installations over a period of 3 years shows that one in six errors result from inadequate storage capacity and yet, little research has been done in the area of storage capacity forecasting that could mitigate these errors. In this paper, we propose Soothsayer, a simulation model that accurately predicts capacity usage by employing 3 techniques: autoregressive and moving-average modeling, clustering and stochastic modeling, and linear regression. Furthermore, our models provide a range of times when the capacity is likely to be reached, rather than a single point estimate, which is more beneficial in capacity planning. We evaluate the accuracy of our model using synthetic data as well as real world data. Our results show that our models outperform the previous piecewise regression method proposed by Chamness when applied to nonlinear datasets, while performing comparably when applied to linear datasets.

The continuing frequency and seriousness of security incidents underline the critical importance of application security. Decentralized information flow control (DIFC), a promising tool for improving application security, gives application developers fine-grained control over security policy and privilege management. DIFC developers can partition much application functionality into untrusted components bound by a kernel- or language-enforced security policy. Unless a (usually smaller and less exposed) trusted component is exploited, the effects of an application compromise are contained by the policy.

Asbestos, a new operating system, provides novel labeling and isolation mechanisms that help contain the effects of exploitable software flaws. Applications can express a wide range of policies with Asbestos's kernel-enforced labels, including controls on interprocess communication and system-wide information flow. A new event process abstraction defines lightweight, isolated contexts within a single process, allowing one process to act on behalf of multiple users while preventing it from leaking any single user's data to others. A Web server demonstration application uses these primitives to isolate private user data. Since the untrusted workers that respond to client requests are constrained by labels, exploited workers cannot directly expose user data except as allowed by application policy. The server application requires 1.4 memory pages per user for up to 145,000 users and achieves connection rates similar to Apache, demonstrating that additional security can come at an acceptable cost.

Asbestos, a new prototype operating system, provides novel labeling and isolation mechanisms that help contain the effects of exploitable software flaws. Applications can express a wide range of policies with Asbestos's kernel-enforced label mechanism, including controls on inter-process communication and system-wide information flow. A new event process abstraction provides lightweight, isolated contexts within a single process, allowing the same process to act on behalf of multiple users while preventing it from leaking any single user's data to any other user. A Web server that uses Asbestos labels to isolate user data requires about 1.5 memory pages per user, demonstrating that additional security can come at an acceptable cost.