Victor A. E. de Farias

b-Coloring graphs with large girth

A b-coloring of a graph is a coloring of its vertices such that every color class contains a vertex that has a neighbor in all other classes. The b-chromatic number of a graph is the largest integer k such that the graph has a b-coloring with k colors. We show how to compute in polynomial time the b-chromatic number of a graph of girth at least 9. This improves the seminal result of Irving and Manlove on trees.

Machine Learning Approach for Cloud NoSQL Databases Performance Modeling

Cloud computing is a successful, emerging paradigm that supports on-demand services with pay-as-you-go model. With the exponential growth of data, NoSQL databases have been used to manage data in the cloud. In these newly emerging settings, mechanisms to guarantee Quality of Service heavily relies on performance predictability, i.e., the ability to estimate the impact of concurrent query execution on the performance of individual queries in a continuously evolving workload. This paper presents a performance modeling approach for NoSQL databases in terms of performance metrics which is capable of capturing the non-linear effects caused by concurrency and distribution aspects. Experimental results confirm that our performance modeling can accurately predict mean response time measurements under a wide range of workload configurations.

Elastic provisioning for cloud databases with uncertainty management

Cloud computing is a compelling, emerging paradigm that supports on-demand services with pay-as-you-go model. It is fundamental for cloud providers to allocate resource quantities suitable to ensure performance while reducing the operational costs related to both overprovisioning and penalties for SLA violations. Performance of cloud services may be unstable because of the environment. Most automatic provisioning techniques lack the capacity to handle the uncertainty of service performance. In this work, we investigate uncertainty management for cloud database elasticity from a probabilistic, performance-driven standpoint, and propose ProDBC, a novel approach for elastic provisioning. ProDBC uses a separate network for profiling in order to build a probabilistic model describing the relationship between workload, resource quantities and the subsequent performance. This model is embedded into a cost function in which the trade-off between infrastructure cost, SLA violation rate and the confidence level (uncertainty) is controlled intuitively. Experimental results obtained with the OLTP Database Benchmark showed provisioning actions taken by ProDBC provides elasticity while limiting the number of SLA violations.

G2P: A Partitioning Approach for Processing DBSCAN with MapReduce

One of the most important aspects to consider when computing large data sets is to distribute and parallelize the analysis algorithms. A distributed system presents a good performance if the workload is properly balanced. It is expected that the computing time is directly related to the processing time on the node where the processing takes longer. This paper aims at proposing a data partitioning strategy that takes into account partition balance and that is generic for spatial data. Our proposed solution is based on a grid model data structure that is further transformed into a graph partitioning problem, where we finally compute the partitions. Our proposed approach is used on the distributed DBSCAN algorithm and it is focused on finding density areas in a large data set using MapReduce. We call our approach G2P (Grid and Graph Partitioning) and we show via massive experiments that G2P presents great quality data partitioning for the distributed DBSCAN algorithm compared to the competitors. We believe that G2P is not only suitable for DBSCAN algorithm, but also to execute spatial join operations and distance based range queries to name to a few.

Towards improvements on the quality of service for multi-tenant RDBMS in the cloud

Cloud computing is a trend of technology aimed at providing on-demand services with payment based on usage. To improve the use of resources, cloud providers adopt multi-tenant approaches, reducing the operation cost of services. Moreover, tenants have irregular workload patterns, impacting in the guarantees of quality of service, mainly due to interference between tenants. This paper proposes an approach to improve quality of service for multi-tenant RDBMS. This approach employs migration techniques of tenants, system monitoring, allocation strategy, forecast approach, and benefits of cloud infrastructure to improve performance and reduce provider cost. We carried out experiments on performance and resource usage in order to evaluate it.

PopRing: A Popularity-aware Replica Placement for Distributed Key-Value Store.

Distributed key-value stores (KVS) are a well-established approach for cloud data-intensive applications, but they were not designed to consider workloads with data access skew, mainly caused by popular data. In this work, we analyze the problem of replica placement on KVS for workloads with data access skew. We formally define our problem as a multi-objective optimization and present the PopRing approach based on genetic algorithm to find a new replica placement scheme. We also use OpenStack-Swift as the baseline to evaluate the performance improvements of PopRing under different configurations. A moderate PopRing configuration reduced in 52% the load imbalance and in 32% the replica placement maintenance while requiring the reconfiguration (data movement) of only 6% of total system data.

Regression based performance modeling and provisioning for NoSQL cloud databases

Abstract Cloud computing is a successful and emerging paradigm that supports on-demand services with pay-as-you-go model. Because of the exponential growth of data, NoSQL databases have been used to manage data in the cloud. In this scenario, it is fundamental for cloud providers guarantee Quality of Service (QoS) by avoiding violations to Service Level Agreement (SLA) contract while reducing the operational costs related to overprovisioning and underprovisioning. In this regard, elastic provisioning mechanisms are employed to maintain QoS by dynamically adding and removing resources to handle workload fluctuations. These mechanisms can also take more accurate provisioning decisions based on performance predictions of the cluster shrinkage and growth. Performance prediction is a challenging task since concurrent access of distributed data can cause non-linear effects on performance. This paper presents a performance modeling approach for NoSQL databases in terms of SLA-based metrics capable of capturing non-linear effects caused by concurrency and distribution aspects. Moreover we present a elastic provisioning strategy that takes advantage on performance models to deliver a reliable resource provisioning. We carried out experiments in order to evaluate our performance modeling and provisioning approaches. The results confirmed that our performance modeling can accurately predict throughput and SLA violations measurements under a wide range of workload settings and also that our elastic provisioning approach can ensure QoS while using resources efficiently.

Towards an Efficient and Distributed DBSCAN Algorithm Using MapReduce

Clustering is a major data mining technique that groups a set of objects in such a way that objects in the same group are more similar to each other than to those in other groups. Among several types of clustering, density-based clustering algorithms are more efficient in detecting clusters with varied density and different shapes. One of the most important density-based clustering algorithms is DBSCAN. Due to the huge size of generated data by the widespread diffusion of wireless technologies and the complexity of big data analysis, new scalable algorithms for efficiently processing such data are needed. In this chapter we are particularly interested in using traffic data for finding congested areas in a city. For this purpose, we developed a new distributed and efficient strategy of DBSCAN algorithm that uses MapReduce to detect dense areas based on the input parameters. We conducted experiments using real traffic data of a brazilian city, Fortaleza, and compared our approach with the centralized and the MapReduce-based approaches. Our preliminary results confirmed that our approach is scalable and more efficient than the other ones. We also present an incremental version of DBSCAN considering the MapReduce version of it.