Distributed Systems -An Algorithmic Approach Di...
Download === https://www.google.com/url?q=https%3A%2F%2Fblltly.com%2F2tDQa7&sa=D&sntz=1&usg=AOvVaw0T2zMDMNDj5os9460brIJv
By leveraging the CAD algorithmic approach, we built VINEA, a prototype implementation and a local virtual network testbed of a Policy-based Virtual Network Embedding Architecture, inspired by our work on the Recursive Internetworking Architecture (RINA) architecture (see e.g., our NSDI 2013 demo). The VINEA prototype code can be found on this github repository.
Applications make extensive use of services offered by distributed platforms ranging from software services to application platforms or mere computational resources. In these cross-domain environments applications may have dependencies on services or resources provided by different domains. A service management solution based on a centrally managed configuration management database (CMDB) is not viable in these environments since CMDB federation does not scale well to many domains. In this paper we propose a distributed configuration management approach by applying standard technologies (e.g., REST services, ATOM feeds) to provide access to and distribution of configuration information. A domain exposes individual configuration items as RESTful web service resources that can be referred to and read by other domains in the context of service management processes. Using this distributed approach, organizations can engage in effective service management practices avoiding the tight integration of CMDBs with their service providers and customers.
Over the last decades, the advancements in measurement, collection, and storage of data have provided tremendous amounts of information. Thus, it has become crucial to extract valuable features and analyze the characteristics of data. As we study more complex systems (e.g. a network of sensors), the relationship between the information in different parts (e.g. measured signals) brings more insight in describing the characteristics of the system. Quantities such as entropy, mutual information, and directed information (DI) can be employed for this purpose. The main theme of this thesis is to study causality between random processes in systems where the instantaneous samples may depend on the history of other processes. We justify utilizing DI to describe the extent of causal influence and provide appropriate tools to estimate this quantity. Additionally, we study properties of the directed information graph, a representation model to demonstrate causal relationships in a network of processes. Although conventional estimation techniques for information-theoretic quantities may suit small systems with low-dimensional data, recent studies acknowledge that these methods may encounter a deterioration in performance when the data is high-dimensional. The estimation techniques proposed in this thesis are aimed to tackle this issue by using a novel approach based on neural networks. A major challenge of this method to estimate DI is to construct appropriate data batches to train the neural network. Thus, we propose a technique using the $k$ nearest neighbors ($k$-NN) algorithm to re-sample the original data. Since DI is characterized with conditional mutual information (CMI) terms, the convergence of our estimators is shown in two steps. First, we prove that the estimation for CMI converges asymptotically to the true value, when samples are independent and identically distributed (i.i.d.). The proof includes a concentration bound for our $k$-NN re-sampling technique. In the next step, the results are extended to the case where samples are allowed to be dependent in time which enables the method to estimate DI. Accordingly, we provide the convergence results for the end-to-end estimation of DI in this scenario. The performance of estimations is investigated in several experiments both with synthetic and real-world data.
We study nonconvex distributed optimization in multi-agent networks with time-varying (nonsymmetric) connectivity. We introduce the first algorithmic framework for the distributed minimization of the sum of a smooth (p
