Alberto Mozo

Testing a Cloud Provider Network for Hybrid P2P and Cloud Streaming Architectures

The number of online real-time streaming services deployed over network topologies like P2P or centralized ones has remarkably increased in the recent years. This has revealed the lack of networks that are well prepared to respond to this kind of traffic. A hybrid distribution network can be an efficient solution for real-time streaming services. This paper contains the experimental results of streaming distribution in a hybrid architecture that consist of mixed connections among P2P and Cloud nodes that can interoperate together. We have chosen to represent the P2P nodes as Planet Lab machines over the world and the cloud nodes using a Cloud providers network. First we present an experimental validation of the Cloud infrastructures ability to distribute streaming sessions with respect to some key streaming QoS parameters: jitter, throughput and packet losses. Next we show the results obtained from different test scenarios, when a hybrid distribution network is used. The scenarios measure the improvement of the multimedia QoS parameters, when nodes in the streaming distribution network (located in different continents) are gradually moved into the Cloud provider infrastructure. The overall conclusion is that the QoS of a streaming service can be efficiently improved, unlike in traditional P2P systems and CDN, by deploying a hybrid streaming architecture. This enhancement can be obtained by strategic placing of certain distribution network nodes into the Cloud provider infrastructure, taking advantage of the reduced packet loss and low latency that exists among its datacenters.

CLUS: Parallel Subspace Clustering Algorithm on Spark

Subspace clustering techniques were proposed to discover hidden clusters that only exist in certain subsets of the full feature spaces. However, the time complexity of such algorithms is at most exponential with respect to the dimensionality of the dataset. In addition, datasets are generally too large to fit in a single machine under the current big data scenarios. The extremely high computational complexity, which results in poor scalability with respect to both size and dimensionality of these datasets, give us strong motivations to propose a parallelized subspace clustering algorithm able to handle large high dimensional data. To the best of our knowledge, there are no other parallel subspace clustering algorithms that run on top of new generation big data distributed platforms such as MapReduce and Spark. In this paper we introduce CLUS: a novel parallel solution of subspace clustering based on SUBCLU algorithm. CLUS uses a new dynamic data partitioning method specifically designed to continuously optimize the varying size and content of required data for each iteration in order to fully take advantage of Spark’s in-memory primitives. This method minimizes communication cost between nodes, maximizes their CPU usage, and balances the load among them. Consequently the execution time is significantly reduced. Finally, we conduct several experiments with a series of real and synthetic datasets to demonstrate the scalability, accuracy and the nearly linear speedup with respect to number of nodes of the implementation.

CogNet: A network management architecture featuring cognitive capabilities

It is expected that the fifth generation mobile networks (5G) will support both human-to-human and machine-to-machine communications, connecting up to trillions of devices and reaching formidable levels of complexity and traffic volume. This brings a new set of challenges for managing the network due to the diversity and the sheer size of the network. It will be necessary for the network to largely manage itself and deal with organisation, configuration, security, and optimisation issues. This paper proposes an architecture of an autonomic self-managing network based on Network Function Virtualization, which is capable of achieving or balancing objectives such as high QoS, low energy usage and operational efficiency. The main novelty of the architecture is the Cognitive Smart Engine introduced to enable Machine Learning, particularly (near) real-time learning, in order to dynamically adapt resources to the immediate requirements of the virtual network functions, while minimizing performance degradations to fulfill SLA requirements. This architecture is built within the CogNet European Horizon 2020 project, which refers to Cognitive Networks.

Can machine learning aid in delivering new use cases and scenarios in 5G

5G represents the next generation of communication networks and services, and will bring a new set of use cases and scenarios. These in turn will address a new set of challenges from the network and service management perspective, such as network traffic and resource management, big data management and energy efficiency. Consequently, novel techniques and strategies are required to address these challenges in a smarter way. In this paper, we present the limitations of the current network and service management and describe in detail the challenges that 5G is expected to face from a management perspective. The main contribution of this paper is presenting a set of use cases and scenarios of 5G in which machine learning can aid in addressing their management challenges. It is expected that machine learning can provide a higher and more intelligent level of monitoring and management of networks and applications, improve operational efficiencies and facilitate the requirements of the future 5G network.

A Fast Iterative Algorithm for Improved Unsupervised Feature Selection

Dimensionality reduction is often a crucial step for the successfulapplication of machine learning and data mining methods. One way toachieve said reduction is feature selection. Due to the impossibilityof labelling many data sets, unsupervised approaches arefrequently the only option. The column subset selection problemtranslates naturally to this purpose, and has received considerableattention over the last few years, as it provides simple linear modelsfor data reconstruction. Existing methods, however, often achieveapproximation errors that are far from the optimum. In this paper wepresent a novel algorithm for column subset selection thatconsistently outperforms state-of-the-art methods in approximationerror. We present a series of key derivations that allow anefficient implementation, making it comparable in speed and in somecases faster than other algorithms. We also prove results that make itpossible to deal with huge matrices, which has strong implications for otheralgorithms of this type in the big data field. We validate our claimsthrough experiments on a wide variety of well-known data sets.

NPEPE: Massive Natural Computing Engine for Optimally Solving NP-complete Problems in Big Data Scenarios

Networks of Evolutionary Processors (NEP) is a bio-inspired computational model defining theoretical computing devices able to solve NP-complete problems in an efficient manner. Networks of Polarized Evolutionary Processors (NPEP) is an evolution of the NEP model that presents a simpler and more natural filtering strategy to simulate the communication between cells. Up to now, it has not been possible to have implementations neither in vivo nor in vitro of these models. Therefore, the only way to analyze and execute NPEP devices is by means of ultra-scalable simulators able to encapsulate the inherent parallelism in their computations. Nowadays, there is a lack of such simulators able to handle the size of non trivial problems in a massively distributed computing environment. We propose as novelty NPEPE, a high scalability engine that runs NPEP descriptions using Apache Giraph on top of Hadoop platforms. Giraph is the open source counterpart of Google Pregel, an iterative graph processing system built for high scalability. NPEPE takes advantage of the inherent Giraph and Hadoop parallelism and scalablity to be able to deploy and run massive networks of NPEPs. We show several experiments to demonstrate that NPEP descriptions can be easily deployed and run using a NPEPE engine on a Giraph+Hadoop platform. To this end, the well known 3-colorability NP complete problem is described as a network of NPEPs and run on a 10 nodes cluster.

Massively Parallel Unsupervised Feature Selection on Spark

High dimensional data sets pose important challenges such as the curse of dimensionality and increased computational costs. Dimensionality reduction is therefore a crucial step for most data mining applications. Feature selection techniques allow us to achieve said reduction. However, it is nowadays common to deal with huge data sets, and most existing feature selection algorithms are designed to function in a centralized fashion, which makes them non scalable. Moreover, some of them require the selection process to be validated according to some target, which constrains their applicability to the supervised learning setting. In this paper we propose as novelty a parallel, scalable, exact implementation of an existing centralized, unsupervised feature selection algorithm on Spark, an efficient big data framework for large-scale distributed computation that outperforms MapReduce when applied to multi-pass algorithms. We validate the efficiency of the implementation using 1GB of real Internet traffic captured at a medium-sized ISP.

Forecasting short-term data center network traffic load with convolutional neural networks

Efficient resource management in data centers is of central importance to content service providers as 90 percent of the network traffic is expected to go through them in the coming years. In this context we propose the use of convolutional neural networks (CNNs) to forecast short-term changes in the amount of traffic crossing a data center network. This value is an indicator of virtual machine activity and can be utilized to shape the data center infrastructure accordingly. The behaviour of network traffic at the seconds scale is highly chaotic and therefore traditional time-series-analysis approaches such as ARIMA fail to obtain accurate forecasts. We show that our convolutional neural network approach can exploit the non-linear regularities of network traffic, providing significant improvements with respect to the mean absolute and standard deviation of the data, and outperforming ARIMA by an increasingly significant margin as the forecasting granularity is above the 16-second resolution. In order to increase the accuracy of the forecasting model, we exploit the architecture of the CNNs using multiresolution input distributed among separate channels of the first convolutional layer. We validate our approach with an extensive set of experiments using a data set collected at the core network of an Internet Service Provider over a period of 5 months, totalling 70 days of traffic at the one-second resolution.

Biased selection for building small-world networks

Small-world networks are currently present in many distributed applications and can be built augmenting a base network with long-range links using a probability distribution. Currently available distributed algorithms to select these long-range neighbors are designed ad hoc for specific probability distributions. In this paper we propose a new algorithm called Biased Selection (BS) that, using a uniform sampling service (that could be implemented with, for instance, a gossip-based protocol), allows to select long-range neighbors with any arbitrary distribution in a distributed way. This algorithm is of iterative nature and has a parameter r that gives its number of iterations. We prove that the obtained sampling distribution converges to the desired distribution as r grows. Additionally, we obtain analytical bounds on the maximum relative error for a given value of this parameter r. Although the BS algorithm is proposed in this paper as a tool to sample nodes in a network, it can be used in any context in which sampling with an arbitrary distribution is required, and only uniform sampling is available. The BS algorithm has been used to choose long-range neighbors in complete and incomplete tori, in order to build Kleinbergs small-world networks. We observe that using a very small number of iterations (1) BS has similar error as a simulation of the Kleinbergs harmonic distribution and (2) the average number of hops with greedy routing is no larger with BS than in a Kleinberg network. Furthermore, we have observed that before converging to the performance of a Kleinberg network, the average number of hops with BS is significantly smaller (up to 14% smaller in a 1000 × 1000 network).

Scalable tag search in social network applications

Emerging social network applications for sharing and collaboration need a way to publish and search a big amount of social objects. Currently, social applications allow the association of a set of user defined keywords, named tags, when publishing these objects, in order to allow searching for them later using a subset of these tags. Commercial systems and recent research community proposals preclude a wide Internet deployment due to the emergence of scalability and hot spot problems in the nodes. We propose T-DHT, an innovative hybrid unstructured-structured DHT based approach, to cope with these high demanding requirements, in a fully scalable, distributed and balanced way. The storage process allows attaching a set of user tags to the stored object and takes at most O(Log(N)) node hops. The tag information attached to the object is stored in a compact way into the node links using a bloom filter, in order to be used later in the search process. The search process allows searching for previously stored objects by means of a tag conjunction and also takes at most O(Log(N)) node hops. The search process is based on DHT typical search combined with an unstructured search algorithm using the tag information previously stored into bloom filters of node links. The simulation results show T-DHT performs in a fully balanced and scalable way, without generating typical hot spot problems even if unbalanced distributions of popular tags are used. Although T-DHT has been devised to build a scalable infrastructure for social applications, it can be applied to solve the more general Peer-to-Peer keyword search problems.