Marta Chinnici

Review on Performance Metrics for Energy Efficiency in Data Center: The Role of Thermal Management

Energy consumption and thermal performance are the two most important tasks in data centers (DCs) facility management. In recent years, to monitor and control their variation several performance metrics were introduced. In this paper an overview on the main important energy and thermal metrics is provided. A critical analysis to investigate mutual relations among metrics was performed, with the aim to clarify some physical aspects regarding the assessment of DC global energy performance.

Measuring energy efficiency in data centers

Energy efficiency in Data Centers (DCs) is currently becoming a topic of increasing importance, considering the rising prices of energy and the expansion of large data sets (Big Data) processing demand. A structured measurement framework that can be used to quantify energy efficiency is required to understand the opportunities for improving energy efficiency in DCs. In other words, a detailed analysis of energy metrics is needed. However, only a small step forward has been made in the measurement of DCs’ energy efficiency in recent years. Therefore, the measurement of energy efficiency in DCs, through a set of globally accepted metrics, is an ongoing challenge. This chapter presents a comprehensive overview of the existing energy, thermal and productivity metrics for DCs and a critical analysis that investigates the intertwined nature of their action areas. The study provides a general methodology that can be used to measure the energy efficiency of DCs through a holistic approach in which the advantages and the disadvantages of existing and emerging metrics are considered critically.

Identifying sparse and dense sub-graphs in large graphs with a fast algorithm

Identifying the nodes of small sub-graphs with no a priori information is a hard problem. In this work, we want to find each node of a sparse sub-graph embedded in both dynamic and static background graphs, of larger average degree. We show that exploiting the summability over several background realizations of the Estrada-Benzi communicability and the Krylov approximation of the matrix exponential, it is possible to recover the sub-graph with a fast algorithm with computational complexity O(N n). Relaxing the problem to complete sub-graphs, the same performance is obtained with a single background. The worst case complexity for the single background is O(n log(n)).

An HPC-Data Center Case Study on the Power Consumption of Workload

With the increasing popularity of Data Center (DCs), the energy efficiency issue is becoming more important than before. Due to their complex nature, the analysis and in particular the measurement of DCs’ energy efficiency is articulated and open issue. Therefore, the analysis of energy efficiency in DCs, through a set of globally accepted metrics, is an ongoing challenge. In particular, the area of productivity metrics is not complete explored and existing proposed metrics none provides a direct measure of the useful work in a DC. To this end, this paper study and analyses the relationship between the power consumption by server’ workload and the relative number of cores used. In details, through the ENEA-HPC’DC facility, we analyse the real data collected during one year to understand the link between workload’ power consumption and cores. In this way, we present to advance beyond the state of the art of the productivity metrics, and in the meantime, a step forward regarding server performance and power management since through the statistical data analysis provides the behaviour of server energy consumption.

Understanding “workload-related” metrics for energy efficiency in Data Center

The measurement of Data Center (DC) energy efficiency is a complicated problem, which depends on its architecture, workload and the environmental conditions, and its estimation has attracted a lot of research. Recently, several metrics were proposed to calculate the energy efficiency in DCs. However, none of the currently proposed metrics provides a direct measure of the useful work in a DC. To this end, this work aims to characterise the energy consumed by different types of server workloads to advance current understanding on the calculation of useful work within a DC. In detail, several measurements of the energy consumption employing different workload configurations were performed to understand the behaviour of energy consumption by each workload category. Workloads were simulated using benchmarks that can provide a preliminary assessment of the workload-related metrics. The Input/Output Operation Per Second (IOPS) parameter, which is a standard performance measurement, was employed in the present analysis. In this paper, the proposed procedure has evaluated in experimental campaigns on the ENEA-C.R. Portici facilities.