AFOLABI, ROTIMI and Covenant University, Theses (2022) ENHANCEMENT OF DATA CENTRE POWER CONSUMPTION THROUGH A PREDICTIVE ALGORITHM. ["eprint_fieldopt_thesis_type_phd" not defined] thesis, Covenant University.
![[img]](http://eprints.covenantuniversity.edu.ng/style/images/fileicons/application_pdf.png) |
PDF
Download (220kB) |
Abstract
Data Centres (DCs) are of paramount importance in the telecommunications industry to meet up with the rapid increase in the demand for telecommunication services. However, the cost of power consumption of a DC accounts for about 80% of the cost incurred in maintaining Data Centres. This situation is further exacerbated in a country like Nigeria where there is highly unstable power supply from the national grid. The unstable power supply leading to increase in the cost of maintaining a DC due to alternative sources of power supply required. Several research projects such as power consumption prediction model and energy consumption optimization have been carried out to reduce the power consumption of Data Centres. However, the existing works suffer from assumption that all the modular that are not carrying traffic will be on idle mode. This generates additional heat and requires a cooling system that consumes extra power compared to when it is completely off. Also, some of the techniques proposed in the literature lack an accurate prediction of the power consumption in Data Centres. This research therefore reduced DC power consumption through a predictive algorithm using Genetic Algorithm (GA) with Kalman Filter (KF) parameters. Data were collected from five different servers in Nigeria, named BSC 13, BSC 14, BSC 15, RNC 05 and RNC 06 using power analyser, clamp meter and thermometer. The historical assessment of data collected were carried out for the DC under study. Two years data (January to December of 2019 and 2020) were collected from the five servers. The data were recorded on an hourly basis for each 357 days in 2019 and 358 in 2020, to obtain a total of 8568 and 8592 samples respectively. All the hourly data measured, and the ones displayed by rectifier Human Machine Interface (HMI) were compared to obtain the percentage error and ascertain the integrity of the data. The data were pre-processed for consistency and the final data used for each year under study, consists of 8400 samples. The final data were divided and categorized into two Datasets. The first dataset was used to create a prediction model, while the second dataset was used for testing. The GA was used to obtain best KF parameters, afterwards KF was used to predict the future power consumption on hourly basis for each day of the week. The proposed model gave low power consumption with accurate prediction when compared with the existing models. Linear Power Consumption Prediction Model (LPCPM) and Adaptive Server Utilization Scheme (ASUS) were also utilized with the assumption that the idle servers are not energised when not required, the performances of these models using different metrics when compared to the existing models in literature demonstrates superiority in terms of cost, power consumption reduction and negligible prediction average absolute error of 0.0025 (0.25%) was obtained.
| Item Type: | Thesis (["eprint_fieldopt_thesis_type_phd" not defined]) |
|---|---|
| Uncontrolled Keywords: | Data Centre, Genetic Algorithm, Kalman Filter, Base Station Controller, Radio Network Controller, Linear Power Consumption Prediction Model, Adaptive Server Utilization Scheme, Power Consumption and Samples. |
| Subjects: | T Technology > T Technology (General) T Technology > TK Electrical engineering. Electronics Nuclear engineering |
| Divisions: | Faculty of Engineering, Science and Mathematics > School of Electronics and Computer Science |
| Depositing User: | AKINWUMI |
| Date Deposited: | 20 Jan 2023 13:54 |
| Last Modified: | 20 Jan 2023 13:54 |
| URI: | http://eprints.covenantuniversity.edu.ng/id/eprint/16540 |
Actions (login required)
 |
View Item |
