博 士 学 位 論 文

博

Doctoral Dissertation

内容の要旨 及び 審査結果の要旨

Dissertation Abstracts and

Summaries of the Dissertation Review Results

第２５号

The Twenty-fifth Issue

平成２７年３月

March, 2015

The University of Aizu

はしがき

博士の学位を授与したので、学位規則（昭和２８年４月１日文部省令第９号）第８条の規定 に基づき、その論文の内容の要旨及び論文審査の結果の要旨をここに公表する。

学位記番号に付した｢甲｣は学位規則第４条第１項（いわゆる課程博士）によるものであるこ とを示す。

Preface

On granting the Doctoral Degree to the individuals mentioned below, abstracts of their theses and the theses review results are herewith publicly announced, in according to the provisions provided for in Article 8 of the Ruling of Degrees (Ministry Of Education Ordinance No.9, enacted on April 1, 1953)

The Chinese character, “甲”, at the beginning of the diploma number represents that an

individual has been granted the degree in accordance with the provisions provided for in

Paragraph 4-1 of the Ruling Of Degrees (what is called “Katei Hakase,” or the Doctoral

Degree granted by the University at which the grantee was enrolled.).

- 1 -

目 次

Contents

掲載順

Order

学位記番号 Diploma No.

学位

Degree

氏名

Name

論文題目

Dissertation Title

頁

Page

１ ^甲

^CI

44

博士（コンピュー タ理工学）

CORTEZ RECINOS, Ruth Patricia

ソフトウェア開発の複雑さを減らすため のサービス指向デザイン手法

Service-Oriented Design Methodology to Reduce Software Development Complexity

2

２ ^甲

^CI

45

博士（コンピュー タ理工学）

BANAGE THENNE GEDARA, Samantha

Kumara

オントロジー学習と文脈認識によるウェ ブサービスクラスタリングの改善

Improving Web Services Clustering through Ontology Learning and Context Awareness

7

３ ^甲

^CI

46

博士（コンピュー タ理工学）

BEN AHMED, Akram

信頼性の高いメッシュ構造のメニーコアネット ワークオンチップ設計のための高スループット アーキテクチャと探索アルゴリズム

High-throughput Architecture and Routing Algorithms Towards the Design of Reliable Mesh-based Many-Core Network-on-Chip Systems

9

４ ^甲

^CI

47

博士（コンピュー タ理工学）

李 鶴

LI, He

ソフトウェア・デファインド・ネットワーキ ングのリソースの最適化

Resource Optimizations in Software Defined Networking

12

５ ^甲

^CI

48

博士（コンピュー タ理工学）

顧 琳

GU, Lin

地理的に分散するデータセンターのコ スト効率的な資源管理

Cost Efficient Resource Management in Geo-distributed Data Centers

15

６ ^乙論博 第

2

博士（コンピュー タ理工学）

鄭 懿

ZHENG, Yi

全心臓モデルによる心室細動のコンピ ュータシミュレーション研究

Computer Simulation Study of Ventricular Fibrillation based on Whole-Heart Model

18

- 2 - Name

氏名

CORTEZ RECINOS, Ruth Patricia

The relevant degree

学位の種類

Doctoral degree (in Computer Science and Engineering)

Number of the diploma of the Doctoral Degree

甲

CI

44

The Date of Conferment

March 20, 2015

27

3

20

Requirements for Degree Conferment

学位授与の要件

Please refer to the article five of “University Regulation on University Degrees”

会津大学学位規程 第５条該当

Dissertation Title

論文題目

Service-Oriented Design Methodology to Reduce Software Development Complexity

ソフトウェア開発の複雑さを減らすためのサービス指向 デザイン手法

Dissertation Review Committee Members

University of Aizu, Prof. VAZHENIN, A.

University of Aizu, Prof. BHALLA, S.

University of Aizu, Prof. KLYUEV, V.

University of Aizu, Senior Associate Prof. PAIK, I.

会津大学教授 ｱﾚｸｻﾝﾀﾞｰ ｳﾞｧｼﾞｪﾆﾝ（主査）

会津大学教授 ｻﾊﾞｼｭ ﾊﾞｰﾗ 会津大学教授 ｳﾞｨﾀﾘｰ ｸﾘｭｴﾌ 会津大学上級准教授 白 寅天

- 3 -

Abstract

Reusability and interoperability of applications and data are of vital importance with the increasing adoption of Cloud Computing, Mobile technologies, and Big data. An enabler to these technologies is Service-Oriented Architecture (SOA) introducing principles, strategies, and patterns to develop loosely coupled, standard -based software components where services are the main building blocks. However, implementing a SOA solution is currently a complex and costly endeavor.

Challenges

One of the main benefits of SOA is to reduce the complexity of enterprise applications once is fully adopted. The architecture is composed in several layers to conceptually organize all the elements involved. It has to deal with Basic services functionality such as capability, interface, quality of services, publication, discovery, and selection. As well as composite services functionality for aggregation, coordination, conformance and semantics. Finally, managing services through their life-cycle, service level agreement (SLA), auditing, monitoring, dynamic provision, and scalability. Therefore, the efforts assembling the necessary laye rs and functionality can be a complex and time consuming task.

Integration approaches such as point to point, are not suitable for SOA due to the direct coupling between client and service, which is difficult to maintain when the application complexity increases such as in an enterprise environment or Cloud. The Enterprise Service Bus (ESB) is a middle layer design pattern to handle the connections between services as one to many, and comprising integration, intelligent routing and composition capabilities , reducing most of the maintenance and integration complexity. However, harnessing all this concepts into SOA enterprise architecture requires assembling together a number of products which are usually costly, and requires specialized IT staff. It means th at special training should be provided to work with new tools, standards, and programming models. To better enable the benefits of Cloud technologies it is necessary that SOA is not only adopted by big size enterprises, but that it becomes the development model of next generation of applications of any type and size of company.

Solution approach

This research proposes a Methodology and design pattern named as Virtual

Model-View-Controller, and a programming approach to reduce software development

complexity of SOA implementation. The methodology is motivated from the traditional

Model-View-Controller (MVC) design pattern that organizes the applications in three main

components. However, the MVC pattern is based on Object Oriented concepts, and it does not

address SOA key elements (services, compositions, and service management). Therefore, this

work proposes an extension of the controller with the Enterprise Service Bus (ESB) pattern

comprising the integration logic, and performing functions for data and pro tocol

- 4 -

transformation, message exchange, and intelligent routing necessary for service composition.

The link between view and model is removed and reorganized within the enhanced controller as a virtual link. Dependency Injection (DI) pattern is introduced in the ESB for service virtualization and workflow compositions. It allows decoupling services from hard -coded dependencies on its collaborators, including expensive initialization logic; facilitating the test of individual services.

The decoupling achieved protects the privacy of the business logic (model) from remote reference access, and reduces the complexity of deploying and assembling software components. This means that the efforts of additional code for integration and wrapping the service to be deployed in a service environment are greatly reduced for the developer.

Moreover, if there is a change in the model, it can be deployed without requiring changes in the view; instead are handled at the controller level.

Research methods and Key contributions.

The first contribution is a novel methodology Virtual -MVC to reduce software development complexity of SOA applications, and its corresponding programming approach. The decoupling of the view from the model allows the programmers to concentrate on buil ding new functionality and services without bothering on how the services will be exposed, consumed, and maintained. The integration logic is outsourced from the developer’s work, and the view and model can be built independently. This means a faster deplo yment time, higher productivity, and facilitates service maintenance with minimum impact to the view.

The second contribution is a development framework based on the Virtual -MVC pattern was implemented using Web services standards, J2EE, and WebSphere app lication server. Several case scenarios were developed for E-Learning and Visual programming, formalizing the development approach.

The programming approach was taught and evaluated in a Software Engineering course at University of Aizu. The evaluation of the methodology shows positive results in complexity reduction and deployment time. Findings of this evaluation demonstrated that the pattern was understandable even for novice programmers (not specialized skills required). Analysis with related MVC programming approaches demonstrated the simplification of integration code, by calculating an index of integration code based on the components required to implement the same Web based application using different programming approaches. Virtual -MVC obtained the lowest ratio.

Finally, the Virtual-MVC programming approach is compared with Service Component

Architecture, a related SOA programming approach. The main outcome of the analysis is the

decoupling between services itself, which in Virtual -MVC does not keep any internal

reference in the business logic minimizing the need of modifying the business logic to

- 5 -

participate in new compositions. The competitive advantages of faster deployment and cost reduction for enterprises are one of the main benefits. However, research environments can benefit from SOA adoption, where small to medium-sized software teams are assembled. The rotation of students is higher than in a company, which is a drawback to advance software implementation. By introducing Virtual-MVC the students can be assigned with a set.

Summaries of the Dissertation Review Results

Reusability and interoperability of applications and data are of vital importance with the increasing adoption of Cloud Computing, Mobile technologies, and Big data. An enabler to these technologies is Service-Oriented Architecture (SOA) introducing principles, strategies, and patterns to develop loosely coupled, standard-based software components where services are the main building blocks. However, implementing a SOA solution is currently a complex and costly endeavor. The thesis focuses on designing a Methodology and design pattern to reduce the complexity of developing Service-Oriented applications.

The first contribution is a novel methodology called as Virtual-MVC, and its correspondent programming approach. The decoupling of the view from the model allows the programmers to concentrate on building new functionality and services without bothering on how the services will be exposed, consumed, and maintained. The integration logic is outsourced from the developer work, and the view and model can be built independently. This means a faster deployment time, higher productivity, and facilitates service maintenance with minimum impact to the view.

The second contribution is a development framework based on the Virtual-MVC pattern. It is implemented using Web services standards, J2EE, and WebSphere application server. Several case scenarios are demonstrated for E-Learning and Visual programming, formalizing the development approach.

Third, the programming approach was taught in a Software Engineering course at University of Aizu.

The evaluation shows a positive result of complexity reduction, and reduced learning time. Findings of this evaluation demonstrated that the pattern was understandable even for novice programmers (not specialized skills required). Analysis with related MVC programming approaches demonstrated the simplification of integration code, by calculating an index of integration code based on the amount of classes required to implement the same Web based application using different programming approaches. Virtual-MVC obtained the lowest ratio.

Finally, the Virtual-MVC programming approach is compared with Service Component Architecture, a

related SOA programming approach. The main outcome of the analysis is the decoupling between

services itself. Virtual-MVC does not keep any internal reference in the business logic minimizing

dependencies and, the need of modifying the logic to participate in new compositions. The competitive

advantages of faster deployment and cost reduction for enterprises are one of the main benefits.

- 6 -

However, the proposed SOA adoption is oriented for research environments, where small to medium-sized software teams are assembled. The rotation of students is higher than in a company, which is a drawback to advance software implementation. By introducing Virtual-MVC the students can be assigned with a set of functionalities instead of the whole application, and it can be constructed by aggregations.

The review committee examined and evaluated the response to questions and new contents included

by the candidate, in detail. State regarding responses to the issued pointed out at the preliminary

dissertation review. During the doctoral dissertation review, the applicant gave clear, explicit and

reasonable answers to the comments, and improved the research work and thesis according to their

suggestions. The candidate revised the thesis according with the preliminary reviewer’s

recommendations. On the whole the members expressed a favorable opinion that the candidate has

acquired an ability to do independent research activity.

- 7 - Name

氏名

BANAGE THENNE GEDARA, Samantha Kumara

The relevant degree

学位の種類

Doctoral degree (in Computer Science and Engineering)

Number of the diploma of the Doctoral Degree

甲

CI

45

The Date of Conferment

March 20, 2015

27

3

20

Requirements for Degree Conferment

学位授与の要件

Please refer to the article five of “University Regulation on University Degrees”

会津大学学位規程 第５条該当

Dissertation Title

論文題目

Improving Web Services Clustering through Ontology Learning and Context Awareness

オントロジー学習と文脈認識によるウェブサービスクラ スタリングの改善

Dissertation Review Committee Members

University of Aizu, Senior Associate Prof. PAIK, I.

(Chief Referee)

University of Aizu, Prof. ZHAO, Q.

University of Aizu, Prof. KLYUEV, V.

University of Aizu, Associate Prof. YEN, N.

会津大学上級准教授 白 寅天（主査）

会津大学教授 趙 強福 会津大学教授 ｳﾞｨﾀﾘｰ ｸﾘｭｴﾌ 会津大学准教授 嚴 昱文

- 8 -

Abstract

With the large number of Web services now available via the Internet, service discovery has become a challenging and time-consuming task. Organizing Web services into similar clusters is a very efficient approach to reducing the search space. A principal issue for clustering is computing the semantic similarity between services, especially one for improving clustering accuracy, the other for context awareness. In this thesis, we present two solutions for similarity calculation. Current approaches use methods such as keyword, information-retrieval (IR) or ontology based methods. These approaches have problems that include discovering semantic characteristics, loss of semantic information and a shortage of high-quality ontologies. In this thesis, first we present a method that adopts ontology learning to generate ontologies via the hidden semantic patterns existing within complex terms. If calculating similarity using the generated ontology fails, it then applies an IR-based method. Further, in the ontology based approach, we propose an approach to identifying the cluster center by combining service similarity with term frequency–inverse document frequency values of service names. As the other issue, current approaches do not consider the domain-specific context in measuring similarity.

Therefore, we propose a context-aware similarity (CAS) method that learns domain context by machine learning to produce models of context for terms retrieved from the Web as the our second approach. The CAS method analyzes the hidden semantics of services within a particular domain. To analyze visually the effect of domain context on the clustering results, CAS based clustering approach applies a spherical associated-keyword-space algorithm. Experimental results show that ontology based approach outperforms comparable existing approaches and CAS based approach works efficiently for the domain-context-aware clustering.

Summaries of the Dissertation Review Results

Web service clustering contributes to service discovery and composition. The applicant’s research focuses on improving quality of service clustering as two aspects: improving accuracy of clustering and inclusion of domain context awareness for service clustering. Firstly, he suggested usage of ontology learning to calculate service similarity that impacts on the accuracy. The approach increases the accuracy by about 10% comparing the existing approaches. Secondly, a novel method for domain awareness of services by machine learning was suggested. While existing methods calculate service similarity based on general context, the method shows far better performance to calculate service similarity by domain context.

During the review, the applicant has very good understanding topics on service clustering and suggests

new and reasonable idea to solve current problem on the issue. He calmly presented his research

contents in English. He understood questions from the referees well and answered them suitably and

politely. All the review committee members agreed that the candidate is qualified for his doctoral

degree considering his thesis, publication record, presentation, and study attitude.

- 9 - Name

氏名

BEN AHMED, Akram

The relevant degree

学位の種類

Doctoral degree (in Computer Science and Engineering)

Number of the diploma of the Doctoral Degree

甲

CI

46

The Date of Conferment

March 20, 2015

27

3

20

Requirements for Degree Conferment

学位授与の要件

Please refer to the article five of “University Regulation on University Degrees”

会津大学学位規程 第５条該当

Dissertation Title

論文題目

High-throughput Architecture and Routing Algorithms Towards the Design of Reliable Mesh-based

Many-Core Network-on-Chip Systems

信頼性の高いメッシュ構造のメニーコアネットワークオ ンチップ設計のための高スループットアーキテクチャと 探索アルゴリズム

Dissertation Review Committee Members

University of Aizu, Prof. BEN, A.

University of Aizu, Prof. MIYAZAKI, T.

University of Aizu, Prof. TSUKAHARA, T.

University of Aizu, Prof. KITAMICHI, J.

University of Aizu, Senior Associate Prof. SAITO, H.

会津大学教授 ﾍﾞﾝ ｱﾌﾞﾃﾞﾗｾﾞｸ（主査）

会津大学教授 宮崎 敏明 会津大学教授 束原 恒夫 会津大学教授 北道 淳司 会津大学上級准教授 齋藤 寛

- 10 -

Abstract

Global interconnects are becoming the principal performance bottleneck for high performance Systems-on-Chips (SoCs). Since the main purpose for these systems is to shrink the size of the chip as smaller as possible while seeking at the same time for more scalability, higher bandwidth, and lower latency. Conventional bus-based-systems are no longer reliable architecture for SoCs due to the lack of scalability and parallelism integration, high latency and power dissipation, and low throughput.

During this last decade, Network-on-Chip (NoC) interconnect has been proposed as a promising solution for future SoC designs. It offers more scalability than the shared-bus based interconnection and allows more processors to operate concurrently.

Despite the higher scalability and parallelism integration offered by NoC over traditional shared-bus based systems, it is still not an ideal solution for future large scale SoCs. This is due to some limitations such as high power consumption, high cost communication, and low throughput. Recently, merging NoC to the third dimension (3D-NoCs) has been proposed to deal with those problems, as it was a solution offering lower power consumption and higher speed.

As 3D-NoC architectures started to show their outperformance and energy efficiency against 2D-NoC systems, questions about their reliability to sustain their performance growth begun to arise. This is mainly due to challenges inherited from both 3D-ICs and NoCs: On one side, the complex nature of 3D-IC fabrics and the continuing shrinkage of semiconductor components. Furthermore, the significant heterogeneity in 3D chips which are likely to mix logic layers with memory layers and even more complex technologies increases the fault's probability in a system. On the other side, the single-point-failure nature of NoC introduces a big concern to their reliability as they are the sole communication medium.

As a result, 3D-NoC systems are becoming susceptible to a variety of faults caused by crosstalk, electromagnetic interferences, impact of radiations, oxide breakdown, and so on. A simple failure in a single transistor caused by one of these factors may compromise the entire system reliability where the failure can be illustrated in corrupted message delivery, time requirements unsatisfactory, or even sometimes the entire system collapse.

In this thesis, we propose 3D-Fault-Tolerant-OASIS (3D-FTO), a robust fault-tolerant 3D-NoC router

architecture endorsed with reliable and graceful routing algorithms. The proposed design handles a

large number of faults in the input-buffer, crossbar, and links (which are the most susceptible

components to faults in 3D-NoC systems) leveraging the inherent structural redundancy in the

architecture to work around errors. Contrary to previous works, the proposed system tolerates multiple

faults in a single crossbar with no considerable performance degradation. In addition, the used

algorithms are always minimal (as long as there exist one minimal path) and with the aid of

Random-Access-Buffer (RAB) mechanism, deadlock-freedom is ensured with no significant area nor

power overhead.

- 11 -

The proposed 3D-FTO system was synthesized using Synopsys Design Compiler at 45nm technology CMOS process technology and its layout is obtained using Cadence SoC Encounter. The evaluation results showed the ability of 3D-FTO to work around different kinds of faults ensuring graceful performance degradation while minimizing the additional hardware complexity and remaining power-efficient.

Summaries of the Dissertation Review Results

In this dissertation, a reliable and high-throughput fault-tolerant 3D-NoC system architecture was proposed, named 3D-Fault-Tolerant-OASIS (3D-FTO). The proposed system is endorsed with

contention-aware and graceful routing algorithms to handle a large number of faults in the input-buffer, crossbar, and links (which are the most susceptible components to faults in 3D-NoC systems).

Contrary to previous works, the proposed system tolerates multiple faults in a single crossbar with no considerable performance degradation. In addition, the used algorithms are always minimal (as long as there exist one minimal path) and with the aid of Random-Access-Buffer (RAB) mechanism,

deadlock-freedom is ensured with no significant area nor power overhead.

The evaluation results showed the ability of 3D-FTO to work around different kinds of faults ensuring graceful performance degradation while minimizing the additional hardware complexity and

remaining power-efficient.

The dissertation has been well organized and well documented with the different necessary

information about the proposed research. In addition, it included a detailed survey about the different conducted works to solve the reliability issues in on-Chip interconnects.

In the preliminary review, the candidate provided explicit and reasonable answers to the comments pointed out by the review committee that aims to improve the understanding of the proposed system and the dissertation quality. In the final review, the candidate addressed all the recommendations accordingly and improved the dissertation according to the committee’s suggestions.

In conclusion, the candidate has fulfilled all the formal requirements for the doctoral degree.

- 12 - Name

氏名

LI, He

李 鶴（リー ホー）

The relevant degree

Doctoral degree (in Computer Science and Engineering)

Number of the diploma of the Doctoral Degree

甲

CI

47

The Date of Conferment

March 20, 2015

27

3

20

Requirements for Degree Conferment

学位授与の要件

Please refer to the article five of “University Regulation on University Degrees”

会津大学学位規程 第５条該当

Dissertation Title

論文題目

Resource Optimizations in Software Defined Networking

ソフトウェア・デファインド・ネットワーキングのリソースの 最適化

Dissertation Review Committee Members

University of Aizu, GUO, S.

University of Aizu, Prof. MIYAZAKI, T.

University of Aizu, Prof. BHALLA, S.

University of Aizu, Prof. BEN, A.

会津大学教授 ｿﾝ ｸﾞｵ（主査）

会津大学教授 宮崎 敏明 会津大学教授 ｻﾊﾞｼｭ ﾊﾞｰﾗ 会津大学教授 ﾍﾞﾝ ｱﾌﾞﾃﾞﾗｾﾞｸ

- 13 -

Abstract

Software-defined network (SDN) is the next generation of networking architecture that is dynamic, manageable, cost-effective, and adaptable, making it ideal for the high-bandwidth, dynamic nature of today's applications.

First, we focus on the resource in the SDN structure. It has been shown that Ternary Content

Addressable Memory (TCAM) which is an important resource to the performance of SDN. However, previous literatures, in point of view on rule placement strategies, are still insufficient to provide high scalability for processing large flow sets with a limited TCAM size. So caching is a brand new method for TCAM management which can provide better performance than rule placement. We propose FDRC, an efficient flow-driven rule caching algorithm to optimize the cache replacement in SDN-based networks. Different from the previous packet-driven caching algorithm, FDRC is characterized by trying to deal with the challenges of limited cache size constraint and unpredictable flows. In particular, we design a caching algorithm with low-complexity to achieve high cache hit ratio by prefetching and special replacement strategy for predictable and unpredictable flows, respectively. By conducting extensive simulations, we demonstrate that our proposed caching algorithm significantly outperforms FIFO and Least Recently Used (LRU) algorithms under various network settings.

Second, we proposed our application, EPM, a novel SDN-based framework for assigning privacy preserving multicast in cloud data center, which is an ordinary network resource optimization with SDN. Since multicast is a very efficient way to reduce duplicate packet transferring in cloud data center, there is impossible to implement privacy ensured multicast in cloud data center with existed limited hardware. EPM apply a high efficient multicast address sharing to support unlimited multicast group with limited network resources. From the result of system evaluation, the limited resource could be allocated efficiently with EPM. Even in IPMC avoided network, EPM is a good complementary to support multicast in virtual machines.

Third, we optimize limited resources for guarantee the security of SDN. We present a secure SDN structure, in which each device is managed by multiple controllers, not just a single as in a traditional manner, with the dynamic and isolated instance provide by the cloud. It can resist Byzantine attacks on controllers and the communication links between controllers and SDN switches. Furthermore, we study a controller minimization problem with security requirement and propose a cost-efficient controller assignment algorithm with a constant approximation ratio. From the experiment result, the secure SDN structure has little impact on the network latency, provide better security than general distributed controller, and the proposed algorithm performs higher efficiency than random assignment.

With the rapid growth of mobile devices and upgrades from 3G to 4G, the subsequent of mobile traffic

requires better management of networking. SDN provides a scalability and flexibility structure with

unprecedented programmability and flow-based paradigm that is ideally for mobile networks. Finally,

- 14 -

we extend our work to the SDN-enabled radio access networks. We propose MoRule, an efficient rule management to optimize the existed SDN design for mobile environment. The challenge of our approach is the design of algorithms that automatically off-load rule processing to overcome resource constraints in SDN devices with moving mobile devices. We design a heuristic algorithm for finding feasible rule placement and adapting traffic overhead induced by rule placement in the facet of traffic changes and mobile devices moves. We demonstrate that our algorithm can find feasible rule

placements better than general SDN placement algorithm.

Summary of the Dissertation Review Result

Software-defined networking (SDN) is the next generation of networking architecture that is dynamic, manageable, cost-effective, and adaptable, making it ideal for the high-bandwidth, dynamic nature of today's applications. This dissertation aims at some fundamental and important issues on resource management and optimization in both SDN switches and controllers. At switch level, it proposes a flow-driven rule caching algorithm to optimize the cache replacement and a rule placement algorithm for SDN-enabled wireless access network with mobile users. At the controller level, it presents a secure SDN structure, in which each switch is managed by multiple controllers. It can resist Byzantine attacks with minimum number of controllers for each switch under the given security requirement.

To the best knowledge of the committee, the dissertation is the first to study these problems with efficient solutions validated by extensive experiments. The candidate published two major journal papers (with one in the top journal IEEE Transactions on Cloud Computing) and three major

conference papers in IEEE ComSoc flagship conferences, all as the first author. The applicant shows

strong knowledge and practical skill of topics related to his research. Dissertation writing and oral

presentation are also excellent.

- 15 - Name

氏名

GU, Lin

顧 琳（リン グー）

The relevant degree

Doctoral degree (in Computer Science and Engineering)

Number of the diploma of the Doctoral Degree

甲

CI

48

The Date of Conferment

March 20, 2015

27

3

20

Requirements for Degree Conferment

学位授与の要件

Please refer to the article five of “University Regulation on University Degrees”

会津大学学位規程 第５条該当

Dissertation Title

論文題目

Cost Efficient Resource Management in Geo-distributed Data Centers

地理的に分散するデータセンターのコスト効率的な資 源管理

Dissertation Review Committee Members

University of Aizu, GUO, S.

University of Aizu, Prof. MIYAZAKI, T.

University of Aizu, Prof. COHEN, M.

University of Aizu, Prof. ZHAO, Q.

会津大学教授 ｿﾝ ｸﾞｵ（主査）

会津大学教授 宮崎 敏明 会津大学教授 ﾏｲｹﾙ ｺｰｴﾝ 会津大学教授 趙 強福

- 16 -

Abstract

The explosive growth of demands on big data processing imposes a heavy burden on computation, storage, and communication in geo-distributed data centers, which hence incurs considerable operational expenditure to data center providers. With the rising numbers data centers all over the world, the electricity consumption and communication cost have been increasing drastically as the main operational expenditure (OPEX) to data centers. Therefore, cost minimization has become an emergent issue for data centers in big data era. The geographical heterogeneity of electricity prices and the hardware device difference motivate us to study the cost minimization problem over

geo-distributed data centers.

We first investigate the promising cost efficient methods by exploring the advantage of dynamic voltage and frequency scaling (DVFS). We exploit the problems of dynamic frequency scaling technique, task placement and data center resizing in geo-distributed data center in order to minimize the cost. We formulate the optimization problem that minimizes OPEX while guaranteeing the quality-of-service, e.g., the expected response time of tasks. Furthermore, an optimal solution is discovered for this formulated problem. Experimental results show that our proposals achieve much higher cost-efficiency than the traditional ones, e.g., resizing scheme.

We next investigate the performance optimization issue of big data services in geo-distributed data centers. Different from conventional cloud services, one of the main features of big data services is the tight coupling between data and computation as computation tasks can be conducted only when the corresponding data is available. As a result, three factors, i.e., task assignment, data placement, and data movement, deeply influence operational expenditure of data centers. In this scenario, we are motivated to study the cost minimization problem via a joint optimization of these three factors for big data services in geo-distributed data centers. We jointly study the data placement, task assignment, data center resizing and routing to minimize the overall operational cost in large-scale geo-distributed data centers for big data applications. We first characterize the data processing process using a two-dimensional Markov chain and derive the expected completion time in closed-form, based on which joint optimization is formulated as a mixed-integer nonlinear programming (MINLP) problem.

To tackle the high computational complexity of solving our MINLP, we linearize it into a

mixed-integer linear programming (MILP) problem. Through extensive experiments, we show that our joint-optimization solution has substantial advantage over the approach by two-step separate optimization. The high efficiency of our proposal is validated by extensive simulation based studies.

Several interesting phenomena are also observed from the experimental results.

Furthermore, with the explosion of big data, processing large numbers of continuous data streams, i.e., big data stream processing (BDSP), has become a crucial requirement for many scientific and

industrial applications in recent years. By offering a pool of computation, communication and storage

resources, public clouds, like Amazon's EC2, are undoubtedly the most efficient platforms to meet the

- 17 -

ever-growing needs of BDSP. Public cloud service providers usually operate a number of

geo-distributed data centers across the globe. Different data center pairs are with different inter-data center network costs due to the different locations and distances. While inter-data center traffic in BDSP constitutes a large portion of a cloud provider's traffic demand over the Internet and incurs substantial communication cost, which may even become the dominant operational expenditure factor.

As the data center resources are provided in a virtualized way, the virtual machines (VMs) for stream processing tasks can be freely deployed onto any data centers, provided that the Service Level Agreement (SLA, e.g., quality-of-information) is obeyed. This raises the opportunity, but also a challenge, to explore the inter-data center network cost diversities to optimize both VM placement and load balancing towards network cost minimization with guaranteed quality-of-information. We first propose a general modeling framework that can transform the VM placement into VM selection placement and describe all representative inter-task relationship semantics in BDSP. Based on our novel framework, we then formulate the overall communication cost minimization problem for BDSP into a MILP problem and prove it to be NP-hard. We then propose a computation-efficient solution based on MILP. The high efficiency of our proposal is validated by extensive simulation based studies.

Summary of the Dissertation Review Result

The explosive growth of big data services imposes a heavy burden on computation, storage, and communication in geo-distributed data centers, which hence incurs considerable operational

expenditure to cloud infrastructure and service providers. Therefore, the study is motivated on the cost minimization problem by managing the resources in geo-distributed data centers. The candidate first studies how to apply dynamic voltage and frequency scaling technique to cut the cost of servers, instead of solely relying on traditional data center resizing. Then the candidate considers the cost minimization problem of big data batch processing with joint consideration of data placement, task assignment and data routing. Finally, the VM placement problem for cloud-based big data streaming processing is investigated under the consideration of task semantics and traffic flow management.

The dissertation is the first to study these three problems. Based on the problem formulations, the

candidate proposes three algorithms with low computational complexities and better performance over

traditional solutions as shown by extensive experiments. The results were published in three major

journal papers (with two IEEE Transactions) and five major conference papers. The candidate shows

strong knowledge and practical skill of topics related to her research. Dissertation writing and oral

presentation are also excellent.

- 18 - Name

氏名

ZHENG, Yi

The relevant degree

Doctoral degree (in Computer Science and Engineering)

Number of the diploma of the Doctoral Degree

論博第

2

The Date of Conferment

March 20, 2015

27

3

20

Requirements for Degree Conferment

学位授与の要件

Please refer to the article five of “University Regulation on University Degrees”

会津大学学位規程 第５条該当

Dissertation Title

論文題目

Computer Simulation Study of Ventricular Fibrillation based on Whole-Heart Model

全心臓モデルによる心室細動のコンピュータシミュレ ーション研究

Dissertation Review Committee Members

University of Aizu, Prof. CHEN, W.

University of Aizu, Prof. PHAM, T.

University of Aizu, Associate Prof. HISADA, Y.

University of Aizu, Associate Prof. ZHU, X.

会津大学教授 陳 文西（主査）

会津大学教授 ﾌｧﾝ ﾄｩｱﾝ ﾄﾞｩｸ 会津大学准教授 久田 泰広 会津大学准教授 朱 欣

- 19 -

Abstract

Mechanisms of ventricular fibrillation (VF) remain poorly understood. The “restitution hypothesis” is considered as a promising one that may lead to a breakthrough in understanding VF. It was

hypothesized that VF organization was associated with action potential duration (APD) restitution and conduction velocity (CV) restitution based on animal experiments by Wu et al. Although this

hypothesis was verified by simulations on a cardiac tissue slab, it is obviously necessary to be further investigated with computer simulations on whole-heart models.

However, so far few simulations have been done to study the effects of synergy of APD restitution (APDR) and CV restitution (CVR) on VF organization. In this dissertation, different

diastolic-interval-dependent APDR and CVR settings were incorporated into the Wei–Harumi whole-heart model and VF was induced to check its responses to alternations of the restitution configurations.

Because a virtual heart does not have spontaneous VF, we have to induce VF before we study VF dynamics. Hence we firstly studied the relationship between cardiac vulnerability to VF induction and APDR slope to check the feasibility of inducing VF in the Wei–Harumi model with

diastolic-interval-dependent APDR. Simulation results suggested that a steeper APDR slope was related with a higher VF-induction risk.

Secondly, we investigated the effects of VF-induction configuration by inducing VF at different heart sites with different induction protocols, and found that the VF organization was neither correlated with induction protocols nor induction sites.

As the major work of this dissertation, we intensively studied effects of synergy of the APDR and CVR on VF dynamics. The conversion from multiple-wavelet VF to ventricular tachycardia (VT) and the degeneration from VT to mother-rotor VF were both successfully simulated in the Wei–Harumi whole-heart model. Through analyses of simulation outcomes, we concluded that it was likely that the conventional VF mechanisms could all be explained by the restitution mechanism. In addition, this simulation work supports the hypothesis that the synergy between the APDR (both its slope and heterogeneity) and CVR contributes to the transitions between multiple-wavelet and mother-rotor VF mechanisms.

T wave is important for studying VF mechanism. We obtained some interesting findings concerning the relationship between APD dispersion and the T wave. Based on these findings, we proposed a hypothesis that the anteroposterior APD gradient contributed to the T-wave genesis.

Summaries of the Dissertation Review Results

Ventricular Fibrillation (VF) is the most malignant arrhythmia that can cause sudden cardiac death in a

- 20 -

very short time. Currently there are only some hypotheses available to explain VF mechanism. This dissertation focuses on investigating the famous “restitution hypothesis” that may bridge the gap between two existing major VF hypotheses: the multiple-wavelet and mother-rotor VF mechanisms.

The work of this dissertation was based on the Wei–Harumi whole-heart model and systematically carried out in three solidly connected steps. Firstly, feasibility of the research was checked, and relationship between action potential duration restitution (APDR) and VF induction was verified. The second step eliminated the possible influence of VF-induction methods on VF patterns and

strengthened the relationship between cardiac restitution and VF organization. The third step satisfactorily simulated different VF patterns and supported the hypothesis that the APDR and CVR contributed to the transitions between multiple-wavelet and mother-rotor VF mechanisms. Based on the simulation work, this dissertation proposed that the APDR heterogeneity might be the true reason to the degeneration from ventricular tachycardia to mother-rotor VF.

In addition, to better understand the VF mechanism, the T-wave mechanism was also studied. Based on the simulation results, the candidate proposed that the anteroposterior APD gradient also played an important role in T-wave genesis.

This dissertation was well prepared and written. It contains abundant contents coming from the sufficient work done by the candidate. He demonstrated competent skill and excellent knowledge in computer modeling and numerical simulation for electrophysiological systems. All comments pointed out in the preliminary review were approvingly solved in the final defense, and quality of the

dissertation was greatly enhanced. He is also skillful in English reading, listening, speaking, and

writing. The candidate has satisfied all the requirements for conferment of the doctoral degree and is

able to independently solve academic problems in biomedical information technology field.

- 21 -

博 士 学 位 論 文

Doctoral Dissertation

内容の要旨 及び 審査結果の要旨

Dissertation Abstracts

and

Summaries of the Dissertation Review Results

第25号

The Twenty-fifth Issue

平成２７年３月

March, 2015

発行 会津大学

〒965-8580 福島県会津若松市一箕町鶴賀

TEL: 0242-37-2600

FAX: 0242-37-2526 THE UNIVERSITY OF AIZU Tsuruga, Ikki-machi Aizu-Wakamatsu City

Fukushima, 965-8580 Japan