Back to search

Aristotle University Of Thessaloniki (AUTH)

Μηχανισμοί εξοικονόμησης ενέργειας σε κατανεμημένα συστήματα μεγάλης κλίμακας

Abstract

dc:description

Energy preservation is one of the basic goals of mankind nowadays. Reasons aremainly environmental and cost. The cost of energy is rising and the damage to theenvironment due to CO2 emissions is getting worse. Computers and large-scalesystems are significant contributors to the global energy consumption. In general, asystem that consumes less energy generates less heat and has longer lifetime. Thus, byreducing energy consumption and heat, we can increase system’s reliability.Processors are major contributors in the energy consumption of a computer system.Thus, two power-saving techniques exist at the processor level, DynamicVoltage/Frequency Scaling (DVFS) and Dynamic Power Management (DPM).According to DVFS, processors can execute tasks at various frequencies dependingon the system’s load. For example, a processor can execute tasks at lower speedswhen it is not necessary for the tasks to be executed at the highest speed. DPM putsidle processors at sleep-state where minimum energy is consumed, only whenprocessors’ computational power is not needed at the time. Significant amounts ofenergy can be saved by exploiting the idle time of processors. Usually processors aredeactivated after some time in the idle state. The issue is when and for how longprocessors are put to sleep-state.In large-scale real-time systems, the workload most times consists of single tasks thatdo not communicate with each other and must be executed within a deadline or elsetheir results are worthless. In this case, DVFS and DPM techniques must ensure thattasks will be executed without violating their deadline. In this PhD thesis, energysaving mechanisms based on DVFS and DPM techniques are proposed, in order tominimize energy consumption and maximize performance. Furthermore, schedulingpolicies that take into account energy consumption are proposed. The processors ofthe computing nodes of large-scale systems are characterized based on theirperformance and energy consumption and scheduling policies take into account thisinformation. Additionally, in this PhD thesis, load balancing techniques are proposedthat take into account both performance and energy consumption of the computingnodes of large-scale systems. Load balancing has been proven to improveperformance by transferring tasks in order to evenly redistribute the load of thesystem. Load balancing techniques proposed in this PhD thesis, take into account thecharacteristics of the computing nodes and their goal is to avoid transferring tasks tomore energy consuming processors since this would affect significantly the totalenergy consumption of the system.Failures are a common phenomenon in large-scale systems. A common practice is toreplace a failed processor with a more “green” energy-saving processor. In this thesis,how these replacements due to failures affect performance and energy consumption ofthe system is studied. A large percentage of tasks submitted to large-scale systems is Bag-of-Tasks (BoT)applications. A BoT is a collection of independent tasks that do not communicate witheach other during execution. They are used in astronomy, Monte Carlo simulations,image processing and database searches. Due to their significance, BoT schedulingand maximization of performance have been studied severely by researchers. In thisPhD thesis, we propose scheduling policies, DVFS mechanisms and load balancingtechniques for this type of applications taking into account energy consumption.In summary, the contribution of this thesis lies in the implementation of newtechniques in order to save energy during the execution of real-time tasks and BoTapplications in large-scale systems without affecting the performance of the system.These mechanisms focus on DVFS and DPM techniques, as well as scheduling andload balancing of tasks in large-scale systems.

Degree

thesis:*
Grantor dc:publisher
Aristotle University Of Thessaloniki (AUTH)
Year dc:date
2016

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Terzopoulos, Georgios
  • Τερζόπουλος, Γεώργιος

Subjects

dc:subject × 18

Rights

dc:rights
Statement dc:rights
  • BY_ND
Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/38507

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Terzopoulos, Georgios; Τερζόπουλος, Γεώργιος. Μηχανισμοί εξοικονόμησης ενέργειας σε κατανεμημένα συστήματα μεγάλης κλίμακας. Aristotle University Of Thessaloniki (AUTH), 2016. http://hdl.handle.net/10442/hedi/38507