Riccardo Pinciroli

Stochastic Analysis of Energy Consumption in Pool Depletion Systems

The evolutions of digital technologies and software applications have introduced a new computational paradigm that involves initially the creation of a large pool of jobs followed by a phase in which all the jobs are executed in systems with limited capacity. For example, a number of libraries have started digitizing their old books, or video content providers, such as YouTube or Netflix, need to transcode their contents to improve playback performances. Such applications are characterized by a huge number of jobs with different requests of computational resources, like CPU and GPU. Due to the very long computation time required by the execution of all the jobs, strategies to reduce the total energy consumption are very important.

Modeling Power Consumption in Multicore CPUs with Multithreading and Frequency Scaling

The rapid growth of energy requirements in large data-center has motivated several research projects focusing on the reduction of power consumption. Several techniques have been studied to tackle this problem, and most of them require simple power models to estimate the energy consumption starting from known system parameters. It has been proven that the CPU is the component of a server that is most responsible for its total power consumption: for this reason several power models focusing on this resource has been developed. However, only a few accounts for standard CPU features like dynamic frequency scaling and hyperthreading, which can have a significant impact on the estimation accuracy. In this paper, we present the results from a set of experiments focusing on these CPU features, and we propose a simple power model able to provide accurate power estimates by taking them into account.

Modeling multiclass task-based applications on heterogeneous distributed environments

The volume of data, one of the five “V” characteristics of Big Data, grows at a rate that is much higher than the increase of ability of the existing systems to manage it within an acceptable time. Several technologies have been developed to approach this scalability issue. For instance, MapReduce has been introduced to cope with the problem of processing a huge amount of data, by splitting the computation into a set of tasks that are concurrently executed. The savings of even a marginal time in the processing of all the tasks of a set can bring valuable benefits to the execution of the whole application and to the management costs of the entire data center. To this end, we propose a technique to minimize the global processing time of a set of tasks, having different service requirements, concurrently executed on two or more heterogeneous systems. The validity of the proposed technique is demonstrated using a multiformalism model that consists of a combination of Queueing Networks and Petri Nets. Application of this technique to an Apache Hive case-study shows that the described allocation policy can lead to performance gains on both total execution time and energy consumption.

Modeling Techniques for Pool Depletion Systems

The evolution of digital technologies and software applications has introduced a new computational paradigm that involves the concurrent processing of jobs taken from a large pool in systems with limited computational capacity. Pool Depletion Systems is a framework proposed to analyze this paradigm where an optimal admission policy for jobs allocation is adopted to improve the performance of the system. Markov analysis and discrete event simulation, two techniques adopted to study Pool Depletion Systems framework, may require a long time before providing results, especially when dealing with complex systems. For this reason, a fluid approximation technique is presented in this chapter; in fact, it can provide results in a very short time, slightly decreasing their accuracy.

Characterization and Evaluation of Mobile CrowdSensing Performance and Energy Indicators

Mobile Crowdsensing (MCS) is a contribution-based paradigm involving mobiles in pervasive application deployment and operation, pushed by the evergrowing and widespread dissemination of personal devices. Nevertheless, MCS is still lacking of some key features to become a disruptive paradigm. Among others, control on performance and reliability, mainly due to the contribution churning. For mitigating the impact of churning, several policies such as redundancy, over-provisioning and checkpointing can be adopted but, to properly design and evaluate such policies, specific techniques and tools are required. This paper attempts to fill this gap by proposing a new technique for the evaluation of relevant performance and energy figures of merit for MCS systems. It allows to get insightson them from three different perspectives: end users, contributors and service providers. Based on queuing networks (QN), the proposed technique relaxes the assumptions of existing solutions allowing a stochastic characterization of underlying phenomena through general, non exponential distributions. To cope with the contribution churning it extends the QN semantics of a service station with variablenumber of servers, implementing proper mechanisms to manage the memory issues thus arising in the underlying process. This way, a preliminary validation of the proposed QN model against an analytic one and an in depth investigation also considering checkpointing have been performed through a case study.

Capacity Planning of Fog Computing Infrastructures for Smart Monitoring

Fog Computing (FC) systems represent a novel and promising generation of computing systems aiming at moving storage and computation close to end-devices so as to reduce latency, bandwidth and energy-efficiency. Despite their gaining importance, the literature about capacity planning studies for FC systems is very limited only considering very simplified technological cases. This paper considers a model for the capacity planning of a FC system for smart monitoring applications. More specifically, this paper considers a FC-based rock collapse forecasting system based on a hybrid wired-wireless architecture deployed in the Swiss-Italian Alps. The system is composed by sensing units deployed on rock faces to gather environmental data and FC-units providing high-performance computing for smart monitoring purposes.