Network Resource Optimization in Cloud Computing Environments (bibtex)
by Athanassios Tsipis
Abstract:
Recent technological advances in cloudification and virtualization already support numerous applications and wide deployment of network devices, e.g., Internet of Things (IoT), and Wireless Sensor Networks (WSNs), that are envisioned to reach the extreme number of 25 billion devices by the end of 2020. The expected incre- ment in human-type and machine-type communications introduce a wide variety of communication characteristics with different requirements regarding data rate, latency, mobility, and reliability. In such environments, to compensate for the huge demands in resources and the growing volumes of data, numerous data centers and devices deployed around the world work in unison to efficiently accommodate ser- vice provisioning. Nevertheless, as the number of users and cloud services increases, key optimization problems emerge and become a priority for cloud service providers who focus on scalability and service availability problems in order to minimize the overall operational cost and eventually increase their profit revenues and public ac- ceptance. In this context, the problems of discovering the number and location of virtual machines that instantiate a cloud service, as well as finding efficient ways for user assignment and dynamic resource allocation, become exceedingly complex problems (NP-hard), since their solutions are affected by a large number of criteria, parameters, and constraints. When also considering the high inherent dynamicity and frequent fluctuations in network traffic and conditions, it becomes abundantly clear that classical or strictly centralized approaches, requiring global knowledge of the topology changes, fail to capture the needs for timely scalability and adapt- ability. Thus, the need for procuring new feasible and low-complexity solutions for service location, resource allocation, and data handling, arises, even if these come at the expense of unguaranteed optimal solutions. The present thesis dissertation studies the preceding problems through a different prism, taking advantage of the benefits offered by distributed computing. As such, it is divided into two major focal points of interest, which are as follows. The first explores location allocation issues in Cloud Gaming, the latter being an ideal representative of highly dynamic cloud computing environments that can be straightforwardly be generalized to capture other cloud application domains. Contrary to general approaches the focus is on the fog/edge layer, for which five decentralized policies, relying solely on local network information, are derived and analyzed. First of all, the ``Gaming Admission Control'' (GAC) is presented, a hy- brid policy for addressing the efficient gaming admission by leveraging elements of multi-server clustered games with cloud/fog elasticity. Simulation results showcase its capabilities for dynamic load balancing. GAC feeds into the next major heuristic policy, namely the ``Utilized Servers First'' (USF), which is proposed for simulta- neously tackling both the maximization of allocated users and the minimization of utilized servers. A comparison against popular alternatives visualizes the superior- ity of the USF. Following, an extension to USF is introduced, suitably called the ``Utilized Servers First with Adaptive Reallocation'' (USFAR), offering higher scal- ability by dynamically reallocating the users based on their gaming preferences, in order to maximize the Quality of Experience (QoE), without degrading the previous targets of USF. Evaluation results depict how USFAR outperforms USF in address- ing the individual objectives. Finally, two scalable policies for reaching an optimal placement of rendering services within fog networks, that effectively minimizes de- ployment, service access, and communication costs, are outlined. Both of them employ simple rules to move, replicate, and consolidated the services. The initial, namely the ``Rendering Service Allocation'' (RSA), addresses the service location adhering to capacity constraints of the renderers. The second, namely the ``QoE- aware Rendering Service Allocation'' (QoERSA), investigates the problem under a different scope, abiding to demands regarding the minimum acceptable QoE of the gamers. Simulations of both policies, verify their expected behavior under various topologies and parameters, and underscore their efficiency in reducing the overall costs. Regarding the second focal point, issues affiliating to data handling optimization in the IoT are studied. Under this scope, the initial attempt is on the proposal of a novel distributed synchronization algorithm for achieving high correlation among the sensed WSN data. Opposing past approaches on the problem, that aim at synchronizing the nodes' clocks, the presented algorithm offers synchronization on a per hop basis of the data measurements. Evaluation results and a comparison against a popular clock synchronization algorithm verify the efficacy in capturing time deviation and reducing the message overhead. The focus is then shifted on drone-based data collection, where the impact of drone flight-route trajectory is analyzed. Extensive simulations with artificial data and data traces show how the message overhead is correlated with the order of rotational symmetry, indicating that higher orders of symmetry for the selected trajectory route shape eventuate to lower transmission overhead and thus reduced energy consumption. Finally, two real-world IoT implementations for the monitoring of traffic and environmental con- ditions are presented, for the research domains of smart cities and smart agriculture respectively. The former, namely the ``Vehicle Tracking System'' (VTS), makes use of low cost IoT micro-controllers, coupled with cloud and fog computing proper- ties, to accurately collect, filter, and process vehicular movement data, in order to document the traffic conditions on the road network. A real-world deployment ex- perimentation underscores the potentiality of the VTS, reporting on its efficacious analysis regarding traffic congestion. The latter, on the other hand, suitably named the ``Fog-assisted Environmental MOnitoring System'' (F.E.MO.S.), showcases the capabilities offered by fog-driven dynamic functionality optimization in terms of re- sponse time adaptation, workload equilibration, and energy preservation. Again, a real-world implementation, even under a wildfire crisis scenario, provides insight re- garding the undeniable benefits of the approach, highlighting the highly adjustable nature of F.E.MO.S., which can autonomously alter its functionality to reflect the necessities for precision minoring and agricultural risk forecasting. Commonplace in all aforementioned contributing approaches of the current thesis refers to the fact that they can be straightforwardly generalized to capture the needs optimization requirements of all cloud edge computing environments. In this regard they are highly scalable, establishing full alignment with the industrial expectations and research directions followed by the cloud systems of the future.
Reference:
Athanassios Tsipis, "Network Resource Optimization in Cloud Computing Environments", Ph.D. Thesis, Ionian University, 2021.
Bibtex Entry:
@phdthesis{tsipis2021thesis,
	abstract = {Recent technological advances in cloudification and virtualization already support numerous applications and wide deployment of network devices, e.g., Internet of Things (IoT), and Wireless Sensor Networks (WSNs), that are envisioned to reach the extreme number of 25 billion devices by the end of 2020. The expected incre- ment in human-type and machine-type communications introduce a wide variety of communication characteristics with different requirements regarding data rate, latency, mobility, and reliability. In such environments, to compensate for the huge demands in resources and the growing volumes of data, numerous data centers and devices deployed around the world work in unison to efficiently accommodate ser- vice provisioning. Nevertheless, as the number of users and cloud services increases, key optimization problems emerge and become a priority for cloud service providers who focus on scalability and service availability problems in order to minimize the overall operational cost and eventually increase their profit revenues and public ac- ceptance. In this context, the problems of discovering the number and location of virtual machines that instantiate a cloud service, as well as finding efficient ways for user assignment and dynamic resource allocation, become exceedingly complex problems (NP-hard), since their solutions are affected by a large number of criteria, parameters, and constraints. When also considering the high inherent dynamicity and frequent fluctuations in network traffic and conditions, it becomes abundantly clear that classical or strictly centralized approaches, requiring global knowledge of the topology changes, fail to capture the needs for timely scalability and adapt- ability. Thus, the need for procuring new feasible and low-complexity solutions for service location, resource allocation, and data handling, arises, even if these come at the expense of unguaranteed optimal solutions. The present thesis dissertation studies the preceding problems through a different prism, taking advantage of the benefits offered by distributed computing. As such, it is divided into two major focal points of interest, which are as follows.
The first explores location allocation issues in Cloud Gaming, the latter being an ideal representative of highly dynamic cloud computing environments that can be straightforwardly be generalized to capture other cloud application domains. Contrary to general approaches the focus is on the fog/edge layer, for which five decentralized policies, relying solely on local network information, are derived and analyzed. First of all, the ``Gaming Admission Control'' (GAC) is presented, a hy- brid policy for addressing the efficient gaming admission by leveraging elements of multi-server clustered games with cloud/fog elasticity. Simulation results showcase its capabilities for dynamic load balancing. GAC feeds into the next major heuristic policy, namely the ``Utilized Servers First'' (USF), which is proposed for simulta- neously tackling both the maximization of allocated users and the minimization of utilized servers. A comparison against popular alternatives visualizes the superior- ity of the USF. Following, an extension to USF is introduced, suitably called the ``Utilized Servers First with Adaptive Reallocation'' (USFAR), offering higher scal- ability by dynamically reallocating the users based on their gaming preferences, in order to maximize the Quality of Experience (QoE), without degrading the previous targets of USF. Evaluation results depict how USFAR outperforms USF in address- ing the individual objectives. Finally, two scalable policies for reaching an optimal placement of rendering services within fog networks, that effectively minimizes de- ployment, service access, and communication costs, are outlined. Both of them employ simple rules to move, replicate, and consolidated the services. The initial, namely the ``Rendering Service Allocation'' (RSA), addresses the service location adhering to capacity constraints of the renderers. The second, namely the ``QoE- aware Rendering Service Allocation'' (QoERSA), investigates the problem under a different scope, abiding to demands regarding the minimum acceptable QoE of the gamers. Simulations of both policies, verify their expected behavior under various topologies and parameters, and underscore their efficiency in reducing the overall costs.
Regarding the second focal point, issues affiliating to data handling optimization in the IoT are studied. Under this scope, the initial attempt is on the proposal of a novel distributed synchronization algorithm for achieving high correlation among the sensed WSN data. Opposing past approaches on the problem, that aim at synchronizing the nodes' clocks, the presented algorithm offers synchronization on a per hop basis of the data measurements. Evaluation results and a comparison against a popular clock synchronization algorithm verify the efficacy in capturing time deviation and reducing the message overhead. The focus is then shifted on drone-based data collection, where the impact of drone flight-route trajectory is analyzed. Extensive simulations with artificial data and data traces show how the message overhead is correlated with the order of rotational symmetry, indicating that higher orders of symmetry for the selected trajectory route shape eventuate to lower transmission overhead and thus reduced energy consumption. Finally, two real-world IoT implementations for the monitoring of traffic and environmental con- ditions are presented, for the research domains of smart cities and smart agriculture respectively. The former, namely the ``Vehicle Tracking System'' (VTS), makes use of low cost IoT micro-controllers, coupled with cloud and fog computing proper- ties, to accurately collect, filter, and process vehicular movement data, in order to document the traffic conditions on the road network. A real-world deployment ex- perimentation underscores the potentiality of the VTS, reporting on its efficacious analysis regarding traffic congestion. The latter, on the other hand, suitably named the ``Fog-assisted Environmental MOnitoring System'' (F.E.MO.S.), showcases the capabilities offered by fog-driven dynamic functionality optimization in terms of re- sponse time adaptation, workload equilibration, and energy preservation. Again, a real-world implementation, even under a wildfire crisis scenario, provides insight re- garding the undeniable benefits of the approach, highlighting the highly adjustable nature of F.E.MO.S., which can autonomously alter its functionality to reflect the necessities for precision minoring and agricultural risk forecasting.
Commonplace in all aforementioned contributing approaches of the current thesis refers to the fact that they can be straightforwardly generalized to capture the needs optimization requirements of all cloud edge computing environments. In this regard they are highly scalable, establishing full alignment with the industrial expectations and research directions followed by the cloud systems of the future.},
	author = {Tsipis, Athanassios},
	date-modified = {2021-01-21 20:20:01 +0200},
	keywords = {stphdthesis,olinet, v-corfu},
	month = {1},
	school = {Ionian University},
	title = {{{Network Resource Optimization in Cloud Computing Environments}}},
	type = {{Ph.D. Thesis}},
	year = {2021}}
Powered by bibtexbrowser