Fabio Diniz Rossi

Performance Evaluation of Container-Based Virtualization for High Performance Computing Environments

The use of virtualization technologies in high performance computing (HPC) environments has traditionally been avoided due to their inherent performance overhead. However, with the rise of container-based virtualization implementations, such as Linux VServer, OpenVZ and Linux Containers (LXC), it is possible to obtain a very low overhead leading to near-native performance. In this work, we conducted a number of experiments in order to perform an in-depth performance evaluation of container-based virtualization for HPC. We also evaluated the trade-off between performance and isolation in container-based virtualization systems and compared them with Xen, which is a representative of the traditional hypervisor-based virtualization systems used today.

E-eco

The high energy consumption of data centers has been a recurring issue in recent research. In cloud environments, several techniques are being used that aim for energy efficiency, ranging from scaling the processors frequency, to the use of sleep states during idle periods and the consolidation of virtual machines. Although these techniques enable a reduction in power consumption, they usually impact application performance. In this paper, we present an orchestration of different energy-savings techniques in order to improve the trade-off between energy consumption and application performance. To this end, we implemented the Energy-Efficient Cloud Orchestrator - e-eco - a management system that acts along with the cloud load balancer deciding which technique to apply during execution. To evaluate e-eco, tests were carried out in a real environment using scale-out applications on a dynamic cloud infrastructure, taking into account transactions per second as a performance metric. In addition to the empirical experiments, we also analyzed the scalability of our approach with an enhanced version of the CloudSim simulator. Results of our evaluations demonstrated that e-eco is able to reduce energy consumption up to 25% compared to power-agnostic approaches at a cost of only 6% of extra SLA violations. When compared to existing power-aware approaches, e-eco achieved the best trade-off between performance and energy-savings. These results showed that our orchestration approach showed a better balance in regard to a more energy-efficient data center with smaller impact on application performance when compared with other works presented in the literature.

MPI-blastn and NCBI-TaxCollector: Improving metagenomic analysis with high performance classification and wide taxonomic attachment

Metagenomic sequencing technologies are advancing rapidly and the size of output data from high-throughput genetic sequencing has increased substantially over the years. This brings us to a scenario where advanced computational optimizations are requested to perform a metagenomic analysis. In this paper, we describe a new parallel implementation of nucleotide BLAST (MPI-blastn) and a new tool for taxonomic attachment of Basic Local Alignment Search Tool (BLAST) results that supports the NCBI taxonomy (NCBI-TaxCollector). MPI-blastn obtained a high performance when compared to the mpiBLAST and ScalaBLAST. In our best case, MPI-blastn was able to run 408 times faster in 384 cores. Our evaluations demonstrated that NCBI-TaxCollector is able to perform taxonomic attachments 125 times faster and needs 120 times less RAM than the previous TaxCollector. Through our optimizations, a multiple sequence search that currently takes 37 hours can be performed in less than 6 min and a post processing with NCBI taxonomic data attachment, which takes 48 hours, now is able to run in 23 min.

On the Impact of Energy-Efficient Strategies in HPC Clusters

Energy-aware management strategies are a recent trend towards achieving energy-efficient computing in HPC clusters. One of the approaches behind those strategies is to apply energy-saving states on idle nodes, alternating them among different sleep states that reflect on many power consumption levels. This paper investigated the way such energy-efficient strategies affected the job turnaround time - the elapsed time between when the job is submitted and when the job is completed, including the wait time as well as the jobs actual execution time - in these clusters. Based on the results we proposed a Best-Fit Energy-Aware Strategy that switches the nodes to a sleep state, depending on the throughput of the resource managers job queue. We simulated the proposed strategy using the SimGrid simulator. Our preliminary results showed a reduction of up to 19% in the overall energy consumption and give us a better understanding of the trade-offs involved in using energy-efficient strategies.

Modeling power consumption for DVFS policies

Power-aware management strategies are a trend towards achieving energy-efficient computing environments. One of the approaches behind those strategies is dynamic frequency and voltage scaling (DVFS). Since frequency adjustments may have a negative impact on system performance, users often have to experiment with these policies to find the optimal configuration for their application and energy reduction goals. While the performance impact can be easily measured by the total execution time of an application, power consumption measurements require additional logging and frequently external equipment. The following paper presents a mathematical model to help users estimate the power consumption of their application when using different DVFS policies. A preliminary evaluation shows that the model has 94% accuracy when compared against real-time measurements.

Modeling and simulation of global and sleep states in ACPI‐compliant energy‐efficient cloud environments

The more large‐scale data centers infrastructure costs increase, the more simulation‐based evaluations are needed to understand better the trade‐off between energy and performance and support the development of new energy‐aware resource allocation policies. Specifically, in the cloud computing field, various simulators are able to predict and measure the behavior of applications on different architectures using different resource allocation policies. Yet, only a few of them have the ability to simulate energy‐saving strategies, and none of them support the complete advanced configuration and power interface (ACPI) specification. ACPI defines a terminology for all possible power states of a machine and their associated power rate. The hardware industry has relied on ACPI to provide up‐to‐date standard interfaces for hardware discovery, configuration, power management, and monitoring, enabling a better understanding of the energy consumption level of different hardware states, referred to as ACPI G‐states, S‐states, and P‐states. In this paper, we improve the modeling and simulation of the ACPI G/S‐states and show not only that these states offer different energy‐saving levels but also that state transitions consume energy. In addition, we model the latency to transit between two states and the effects on the turnaround time when the transitions are not performed conservatively. Furthermore, the equations provide essential information to quantify the trade‐off between energy consumption and performance and assist in the analysis/decision on which strategy fits better in the environment and how it could be refined. Our expanded energy model was implemented in CloudSim and validated with simulation‐based experiments with a very high level of accuracy, with a standard deviation of at most 6%. Copyright

Performance Evaluation of Virtualization Technologies for Databases in HPC Environments

Modern database systems are applications that have a variable workload demanding a great share of the available resources during peak periods. To avoid sub utilization of resources during periods of low load, virtualization technology allows this type of application to be provisioned dynamically. This technique has improved in the last decades and new approaches are emerging to improve performance, like operating system level database virtualization. This paper analyzes the characteristics of this approach compared to traditional virtualization techniques.

Improving EDP in multi-core embedded systems through multidimensional frequency scaling

Energy saving management in multi-core embedded environments has been a challenge for designers. To achieve energy efficiency, most studies consider dynamic frequency scaling on one hardware component only, such as processor or memory — which will most likely also affect performance. This work proposes the use of frequency scaling considering the three most important hardware components altogether: processors, L2 cache, and RAM; seeking for the best set of frequencies for each one of them to improve the Energy-Delay Product (EDP), depending on the applications behavior. Therefore, this work addresses multidimensional frequency scaling for multi-core embedded systems. By evaluating different frequency levels, we show that the EDP can be improved in up to 46.4% when compared to the standard way that the frequencies are configured.

Performance and accuracy trade-off analysis of techniques for anomaly detection in IoT sensors

IoT environments are typically composed of hundreds of geographically distributed sensors. Usually, these sensors are not physically protected from unauthorized access, which makes them vulnerable to exploitation where they can be manipulated to send incorrect data. The identification of such compromised sensors can be helpful in the process of exclusion or verification by administrators. To perform the detection of anomalous sensors, several algorithms can be used. However, based on the algorithm used, this evaluation may be delayed or can be inaccurate. Therefore, to detect sensors with different behavior compared to others, we evaluated the trade-off between performance and accuracy of different anomalies detection algorithms. The results showed that Mahalanobis Distance could improve the trade-off between detecting multiple anomalous sensors at execution time and accuracy to avoid false-positives.

Dynamic Network Bandwidth Resizing for Big Data Applications

Big Data concerns processing of large volumes of digital data with high velocity and variety. Big Data technologies allow the analysis of data in real time, which is critical for various eScience applications. In order to meet the growing demand of Big Data applications, the infrastructures must be flexible enough to adapt to the characteristics of the applications. Most of the solutions presented in the literature to support Big Data applications focus on scaling processors and memory to handle a variable demand from applications. In a complementary way, this article targets the problem of adapting the network bandwidth to the amount of data to be transferred to and from the applications in order to improve the performance of the applications. For this purpose, we propose the use link aggregation protocol along with Software-Defined Network capabilities for management of the network flow. Results showed that the proposed approach improves the applications performance by up to 33%.