Altino M. Sampaio

Optimizing Energy-Efficiency in High-Available Scientific Cloud Environments

Virtualization technologies empower construction of flexible computing environments, promising an opportunity for energy and resource cost optimization, while enhancing system availability and achieving high performance. A crucial requirement for effective consolidation is to be able to efficiently utilize system resources for high-availability computing, and energy-efficiency optimization, so as to reduce operational costs and carbon footprints to the environment. In this work, we propose a consolidation technique to improve the performance of energy- and reliability-aware scheduling algorithms. For that, we carefully tune an energy optimization mechanism, which detects energy optimizing opportunities, and executes power- and failure-aware decision making algorithms to readjust virtual-to-physical mappings. We conduct simulations injecting synthetic jobs which characteristics follow the last version of the Google Cloud trace logs. The results indicate that our strategy improves work per Joule ratio in about 9.7%, as well working-efficiency in almost 15.6%, maintaining similar levels of completion jobs.

PIASA: A power and interference aware resource management strategy for heterogeneous workloads in cloud data centers

Abstract Cloud data centers have been progressively adopted in different scenarios, as reflected in the execution of heterogeneous applications with diverse workloads and diverse quality of service (QoS) requirements. Virtual machine (VM) technology eases resource management in physical servers and helps cloud providers achieve goals such as optimization of energy consumption. However, the performance of an application running inside a VM is not guaranteed due to the interference among co-hosted workloads sharing the same physical resources. Moreover, the different types of co-hosted applications with diverse QoS requirements as well as the dynamic behavior of the cloud makes efficient provisioning of resources even more difficult and a challenging problem in cloud data centers. In this paper, we address the problem of resource allocation within a data center that runs different types of application workloads, particularly CPU- and network-intensive applications. To address these challenges, we propose an interference- and power-aware management mechanism that combines a performance deviation estimator and a scheduling algorithm to guide the resource allocation in virtualized environments. We conduct simulations by injecting synthetic workloads whose characteristics follow the last version of the Google Cloud tracelogs. The results indicate that our performance-enforcing strategy is able to fulfill contracted SLAs of real-world environments while reducing energy costs by as much as 21%.

Energy-Efficient and SLA-Based Resource Management in Cloud Data Centers

Abstract Nowadays, cloud data centers play an important role in modern Information Technology (IT) infrastructures, being progressively adopted in different scenarios. The proliferation of cloud has led companies and resource providers to build large warehouse-sized data centers, in an effort to respond to costumers demand for computing resources. Operating with powerful data centers requires a significant amount of electrical power, which translates into more heat to dissipate, possible thermal imbalances, and increased electricity bills. On the other hand, as data centers grow in size and in complexity, failure events become norms instead of exceptions. However, failures contribute to the energy waste as well, since preceding work of terminated tasks is lost. Therefore, todays cloud data centers are faced with the challenge of reducing operational costs through improved energy utilization while provisioning dependable service to customers. This chapter discusses the causes of power and energy consumption in data centers. The advantages brought by cloud computing on the management of data center resources are discussed, and the state of the art on schemes and strategies to improve power and energy efficiency of computing resources is reviewed. A practical case of energy-efficient and service-level agreement (SLA)-based management of resources, which analyzes and discusses the performance of three state-of-the-art scheduling algorithms to improve energy efficiency, is also included. This chapter concludes with a review of open challenges on strategies to improve power and energy efficiency in data centers.

A Comparative Cost Study of Fault-Tolerant Techniques for Availability on the Cloud

The success of ever growing warehouse-sized Cloud data centers built to respond to the increasing demand for computing resources depends on the ability to provide reliability and availability at scale. In order to provide dependable and secure systems and services, one needs to implement security controls capable of avoiding, coping and recovering from failures. However, dependability and security of services at all cost is not a solution for Cloud providers. In this paper, two state-of-the-art fault-tolerance techniques are compared in terms of availability of services to consumers, and energy costs to Cloud providers. The results have shown that proactive fault-tolerance technique outperforms traditional redundancy in terms of cost to Cloud users, while providing available compute environments and services to consumers.

Distributed Prime Sieve in Heterogeneous Computer Clusters

Prime numbers play a pivotal role in current encryption algorithms and given the rise of cloud computing, the need for larger primes has never been so high. This increase in available computation power can be used to either try to break the encryption or to strength it by finding larger prime numbers. With this in mind, this paper provides an analysis of different sieve implementations that can be used to generate primes to near 264. It starts by analyzing cache friendly sequential sieves with wheel factorization, then expands to multi-core architectures and ends with a cache friendly segmented hybrid implementation of a distributed prime sieve, designed to efficiently use all the available computation resources of heterogeneous computer clusters with variable workload and to scale very well in both the shared and distributed memory versions.

A Study on Cloud Cost Efficiency by Exploiting Idle Billing Period Fractions

In most of the current commercial Clouds, resources are billed based on a time interval equal to one hour, as is the case of virtual machine (VM) instances on Amazon EC2. Such time interval is usually long, and yet the user has to pay for the whole last hour, even if he/she has only used a fraction of it, contradicting the pay-as-you-go model of Clouds. In this paper, we analyse the advantages of adopting alternative scheduling policies that exploit idle last time intervals, in terms of service cost to Cloud users and operating costs to Cloud providers. Using a real-life astronomy workflow application, constrained by user-defined Deadline and Budget quality of service (QoS) parameters, a set of online state-of-the-art-based scheduling algorithms try different execution and resource provisioning plans. Our results show that exploitation of partially idle last time intervals can reduce the cost of service to the end user, and augments providers competitiveness up to 21.6% through energy efficiency improvement and consequent lowering of operational costs.

Parallel Algorithms for Multirelational Data Mining: Application to Life Science Problems

Data Mining (DM) algorithms are able to construct models from available data that can be very useful for both business and science. However, a powerful representation language is required to express the highly complex models that stem from structured data. Multirelational algorithms can then take advantage of this representation for both data and models. The drawback is that for very large or highly complex domains multirelational algorithms may require long running times. This problem can be substantially reduced using parallel implementations. In this chapter, we present a survey on parallel approaches to run Inductive Logic Programming (ILP), a flavor of multirelational algorithms. We also analyze different scheduling approaches for those implementations and describe two applications where the proposed approaches may be very useful.