Skip to Main content Skip to Navigation

Scheduling in cloud data center powered by renewable energy only with mixed phases-based workload

Abstract : Due to the increase of cloud, web-services and high performance computing demands all over the world, datacenters are now known to be one of the biggest actors when talking about energy consumption. In 2006 alone, datacenters were responsible for consuming 61.4 billion kWh in the United States. When looking at the global scenario, datacenters are currently consuming more energy than the entire United Kingdom, representing about 1.3\% of world's electricity consumption, and being even called the factories of the digital age. Supplying datacenters with clean-to-use renewable energy is therefore essential to help mitigate climate change. The vast majority of cloud provider companies that claim to use green energy supply on their datacenters consider the classical grid, and deploy the solar panels/wind turbines somewhere else and sell the energy to electricity companies, which incurs in energy losses when the electricity travels throughout the grid. Even though several efforts have been conducted at the computing level in datacenters partially powered by renewable energy sources, the scheduling considering on site renewable energy sources and its variations, without connection to the grid can still be widely explored. Since energy efficiency in datacenters is directly related to the resource consumption of the computing nodes, performance optimization and an efficient load scheduling are essential for energy saving. Today, we observe the use of cloud computing as the basis of datacenters, either in a public or private fashion. The main particularity of our approach is that we consider a power envelope composed only by renewable energy as a constraint, hence with a variable amount of power available at each moment. The scheduling under this kind of constraint becomes more complex: without further checks, we are not ensured that a running task will run until completion. We start by addressing the IT load scheduling of batch tasks, which are characterized by their release time, due date and resource demand, in a cloud datacenter while respecting the aforementioned power envelope. The data utilized for the batch tasks comes from datacenter traces, containing CPU, memory and network values. The power envelopes considered, represent an estimation which would be provided by a power decision module and is the expected power production based on weather forecasts. The aim is to maximize the Quality of Service with a variable constraint on electrical power. Furthermore, we explore a workload composed by batch and services, where the resources consumption varies over time. The traces utilized for the service tasks originate from business critical datacenter. In this case we rely on the concept of phases, where each significant resource change in the resources consumption constitutes a new phase of the given task. In this task model phases could also receive less resources than requested. The reduction of resources can impact the QoS and consequently the datacenter profit. In this approach we also include the concept of cross-correlation to evaluate where to place a task under a power curve, and what is the best node to place tasks together (i.e. sharing resources). Finally, considering the previous workload of batch tasks and services, we present an approach towards handling unexpected events in the datacenter. More specifically we focus on IT related events such as tasks arriving at any given time, demanding more or less resources than expected, or having a different finish time than what was initially expected. We adapt the proposed algorithms to take actions depending on which event occurs, e.g. task degradation to reduce the impact on the datacenter profit.
Document type :
Complete list of metadata

Cited literature [139 references]  Display  Hide  Download
Contributor : Abes Star :  Contact
Submitted on : Tuesday, May 26, 2020 - 11:17:27 PM
Last modification on : Wednesday, November 3, 2021 - 7:17:42 AM


Version validated by the jury (STAR)


  • HAL Id : tel-02628518, version 1


Gustavo Rostirolla. Scheduling in cloud data center powered by renewable energy only with mixed phases-based workload. Databases [cs.DB]. Université Paul Sabatier - Toulouse III, 2019. English. ⟨NNT : 2019TOU30160⟩. ⟨tel-02628518⟩



Les métriques sont temporairement indisponibles