1 Approaches to Minimize Power Consumption of Computation NetworkLarysa Globa1, Oleksandr Stepurin2 1. National Technical University of Ukraine “KPI”, UKRAINE, Kyiv, Industrialnyi alley 2, 2. National Technical University of Ukraine “KPI”, UKRAINE, Kyiv, Industrialnyi alley 2, CADSM 2015, February, 2015, Polyana-Svalyava (Zakarpattya), UKRAINE
2 Abstract In this paper it was reviewed several existing investigations about minimization power consumption of data centers. It was suggested several approaches of task distribution in computing networks to minimize power consumption and management system for such networks.
3 I. Introduction Power consumption increasing is slightly important problem in the major spheres of life Consistently decreasing power resources and their expensiveness forces implementation of green technologies and save as much energy as possible by minimizing it usage. IT technologies also face this issue due to permanent increase of data amount, requirements to calculations accuracy and wide interest to internet and IT technology.
4 Review of Existing InvestigationsResearchers from Dresden technical university investigated power consumption of servers during virtual machines migration. It was shown on practice and confirmed by experiments that movement of VMs and shutting down unused physical servers will help to minimize power consumption of system.
5 Review of Existing InvestigationsVirtual machine migration time investigation. Researches held quite number of experiment and built dependency, which shows influence of virtual machine load on migration time. Summary time of migration of several virtual machines could be reduced if migration will be done one by one in order, which depends on VM’s load type.
6 Approaches to Minimize Power ConsumptionWays to reduce power consumption: Continuously hardware improvement Using the most modern cooling systems Optimization of task scheduling mechanism to minimize power consumption
7 Approaches to Minimize Power ConsumptionThe main goal of this paper is to describe several approaches of task distribution in computing networks to minimize power consumption based on statistical data and live information of data center load. Using statistics information it is possible to get dependency of system behavior in current moment of time and make a forecast for nearest period of time in future for managing computation process to reach maximal energy efficiency.
8 Approaches to Minimize Power Consumption
9 Approaches to Minimize Power ConsumptionInitial data which exist before starting investigation and optimization are: Number of nodes in data center Hardware parameters of each node Hardware requirements of each incoming task on the grid. Dependency of power consumption from load Taking into account all describe information, a classical minimization task could be written Eq. 1: 𝑖=1 𝑀 𝑃 𝑖 𝐿 ∗𝑡→𝑚𝑖𝑛
10 The main difficulty of it’s resolution it is unknown time required to resolve each task on any node in the data center. To solve mentioned uncertainty below ways could be chosen: Use probability characteristics for time of execution based on historical data. Assume that task executes quite long time (more than 1 hour). Divide incoming tasks on several types based on their average execution time (using historical data) and resources consumption.
11 Conclusion In the nearest future it is planned to continue researches in described area. It is planned to develop mathematical model for several described approaches and execute number of experiments using developed models. The main goal is to find the most suitable model for real life. Depending on selected model distribution system which will realize mathematical model will be created. As a next step a real experiment to prove that developed system is energy and cost efficient will be held.
12 Conclusion Described approaches have number of benefits for users of end system data centers. First of all it could provide significant cost saving and energy saving which is quite important nowadays. Also it gives opportunity to redistribute tasks between nodes to reach optimal utilization from power efficiency point of view.
13 Conclusion Such mechanisms could be used in educational centers, institutes, data centers used for high performance calculations and virtualized data centers. System could be quite useful in case of power outage in data center when data center is switched to batteries. In such cases the most critical is to keep running applications which have significant impact as long as it is possible. System will allow to redistribute services between nodes in the most optimal way from power saving point of view and turn off unused machines and terminate non critical services. Such approach will help to increase vitality of data center.