Section: New Results
Energy-aware computing
Participants : Emile Cadorel, Hélène Coullon, Adrien Lebre, Thomas Ledoux, Jean-Marc Menaud, Jonathan Pastor, Dimitri Saingre, Yewan Wang.
Energy consumption is one of the major challenges of modern datacenters and supercomputers. Our works in Energy-aware computing can be categorized into two subdomains: Software level (SaaS, PaaS) and Infrastructure level (IaaS). At Software level, we worked on the general Cloud applications architectures and more recently on BlockChain-based solutions. At Infrastructure level, we worked this year on two directions: (i) investigating the thermal aspects in datacenters, and (ii) analyzing the energy footprint of geo-distributed plateforms.
In [11], the scheduling of heterogeneous scientific workflows while minimizing the energy consumption of the cloud provider is tackled by introducing a deadline sensitive algorithm. Scheduling workflows in a cloud environment is a difficult optimization problem as capacity constraints must be fulfilled additionally to dependencies constraints between tasks of the workflows. Usually, work around the scheduling of scientific workflows focuses on public clouds where infrastructure management is an unknown black box. Thus, many works offer scheduling algorithms designed to select the best set of virtual machines over time, so that the cost to the end user is minimized. This paper presents the new v-HEFT-deadline algorithm that takes into account users deadlines to minimize the number of machines used by the cloud provider. The results show the real benefits of using our algorithm for reducing the energy consumption of the cloud provider.
In [25], over the last year, both academic and industry have increase their work on blockchain technologies. Despite the potential of blockchain technologies in many areas, several obstacles are slowing down their development. In addition to the legal and social obstacles, technical limitations now prevent them from imposing themselves as a real alternative to centralised services. For example, several problems dealing with the scalability or the energy cost have been identified. That's why, a significant part of this research is focused on improving the performances (latency, throughput, energy footprint, etc.) of such systems. Unfortunately, Those projects are often evaluated with ad hoc tools and experimental environment, preventing reproducibility and easy comparison of new contribution to the state of the art. As a result, we notice a clear lack of tooling concerning the benchmarking of blockchain technologies. To the best of our knowledge only a few tools address such issues. Those tools often relies on the load generation aspect and omit some other important aspect of benchmark experiments such as reproducibility and the network emulation. We introduce BCTMark, a general framework for benchmarking blockchain technologies in an emulated environment in a reproductible way.
In [18], we present a deep evaluation about the power models based on CPU utilization. The influence of inlet temperature on models has been especially discussed. According to the analysis, one regression formula by using CPU utilization as the only indicator is not adequate for building reliable power models. First of all, Workloads have different behaviors by using CPU and other hardware resources in server platforms. Therefore, power is observed to have high dispersion for a fixed CPU utilization, especially at full workload. At the same time, we also find that, power is well proportional to CPU utilization within the execution of one single workload. Hence, applying workload classifications could be an effective way to improve model accuracy. Moreover, inlet temperature can cause surprising influence on model accuracy. The model reliability can be questioned without including inlet temperature data. In a use case, after including inlet temperature data, we have greatly improved the precision of model outputs while stressing server under three different ambient temperatures.
In [18], our physical experiments have shown that even under the same conditions, identical processors consume different amount of energy to complete the same task. While this manufacturing variability has been observed and studied before, there is lack of evidence supporting the hypotheses due to limited sampling data, especially from the thermal characteristics. In this article, we compare the power consumption among identical processors for two Intel processors series with the same TDP (Thermal Design Power) but from different generations. The observed power variation of the processors in newer generation is much greater than the older one. Then, we propose our hypotheses for the underlying causes and validate them under precisely controlled environmental conditions. The experimental results show that, with the increase of transistor densities, difference of thermal characteristics becomes larger among processors, which has non-negligible contribution to the variation of power consumption for modern processors. This observation reminds us of re-calibrating the precision of the current energy predictive models. The manufacturing variability has to be considered when building energy predictive models for homogeneous clusters.
In [3], we propose a model and a first implementation of a simulator in order to compare the energy footprint of different cloud architectures (single sites vs fully decentrlaized). Despite the growing popularity of Fog/Edge architectures, their energy consumption has not been well investigated yet. To move forward on such a critical question, we first introduce a taxonomy of different Cloud-related architectures. From this taxonomy, we then present an energy model to evaluate their consumption. Unlike previous proposals, our model comprises the full energy consumption of the computing facilities, including cooling systems, and the energy consumption of network devices linking end users to Cloud resources. Finally, we instantiate our model on different Cloud-related architectures, ranging from fully centralized to completely distributed ones, and compare their energy consumption. The results validates that a completely distributed architecture, because of not using intra-data center network and large-size cooling systems, consumes less energy than fully centralized and partly distributed architectures respectively. To the best of our knowledge, our work is the first one to propose a model that enables researchers to analyze and compare energy consumption of different Cloud-related architectures.