2014 papers by author ieee - グリッド協議会 a framework for dynamic configuration of cloud...

13
IEEE CloudCom2014 参加報告 筑波大学 システム情報系 阿部洋丈 [email protected] Singapore 2014 nternaonal Conference on Clou

Upload: truongngoc

Post on 29-Jun-2019

220 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: 2014 PAPERS BY AUTHOR IEEE - グリッド協議会 A Framework for Dynamic Configuration of Cloud Services Duc-Hung Le, Hong-Linh Truong, Georgiana Copil, Stefan Nastic, Schahram

IEEE CloudCom2014 参加報告

筑波大学 システム情報系 阿部洋丈 [email protected]

CONFERENCE INFORMATION

PAPERS BY SESSION

PAPERS BY AUTHOR

SEARCH

GETTING STARTED

TRADEMARKS

Singapore2014

6th IEEE International Conference on Cloud Computing Technology and Science

Organized by

15-18 December 2014 • Singapore

Page 2: 2014 PAPERS BY AUTHOR IEEE - グリッド協議会 A Framework for Dynamic Configuration of Cloud Services Duc-Hung Le, Hong-Linh Truong, Georgiana Copil, Stefan Nastic, Schahram

IEEE CloudCom2014 概要 ▪  正式名称: 6th IEEE International Conference on Cloud Computing Technology and Science

▪  日程: 2014年12月15日~18日

▪  開催地: Nanyang Technological University (NTU), Singapore ▪  二大有力大学のうちの一つ。シンガポールの西に位置 ▪  会場の Nanyang Exective Center(右図)

▪  IEEE CloudComの歴史 ▪  北京(‘09) → Indianapolis → Athens → 台北 → Bristol, UK → Singapore(‘14) → Vancouver (‘15)

Page 3: 2014 PAPERS BY AUTHOR IEEE - グリッド協議会 A Framework for Dynamic Configuration of Cloud Services Duc-Hung Le, Hong-Linh Truong, Georgiana Copil, Stefan Nastic, Schahram

投稿状況・参加状況 ▪  投稿数: “over 300” ▪  54件がメイン会議に採択。採択率は18% 。そのうち31件は short paper としての採択(つまり full paper は 7.7%!)

▪  日本からの採択は6件 ▪  東工大(+IBM)、慶応大(+農工大)、NTTドコモ、NII、産総研 (+ モンクット王 工科大学(タイ))、日立(+大阪大)

▪  メイン会議に採択されなかった論文の一部は poster presentation や PhD consortium paper, EIC workshop paper として採択 ▪  Poster: 27件、PhD: 5件、EIC WS: 22件 。ここまで含めると採択率36%

▪  参加者: 二百人程度?(目測) ▪  中国~東南アジア~中東からの参加者が比較的多い印象

Page 4: 2014 PAPERS BY AUTHOR IEEE - グリッド協議会 A Framework for Dynamic Configuration of Cloud Services Duc-Hung Le, Hong-Linh Truong, Georgiana Copil, Stefan Nastic, Schahram

メイン会議のプログラム構成 ▪  基調講演 [×5] ▪  通常セッション [×19]

▪  Industry セッション [×2] ▪  PhD consortium [×1] ▪  パネルディスカッション [×1] ▪  チュートリアル [×4] ▪  Reception, River Cruise, Banquet ▪  右図: River Cruise で撮ったマーライオン(真正面)

•  Architecture [×4] •  Services and Applications [×4] •  Big Data [×2] •  HPC on Cloud [×2]

•  Mobile on Cloud [×2] •  Security and Privacy [×2] •  Virtualization [×2] •  Autonomicity & Accountability [×1]

Page 5: 2014 PAPERS BY AUTHOR IEEE - グリッド協議会 A Framework for Dynamic Configuration of Cloud Services Duc-Hung Le, Hong-Linh Truong, Georgiana Copil, Stefan Nastic, Schahram

基調講演 ▪  Building Big Data Processing Systems under Scale-out Computing Model ▪  Prof. Xiaodong Zhang (The Ohio State University, USA) ▪  Apache Hive の最適化の話。Data storage structure, query planning, query execution を改良。

▪  Cloud-Based Systems of Insight ▪  Dr. Hui Lei (IBM Thomas J. Watson Research Center) ▪  IBMの提唱している ”Infrastructure Matters” の一部を構成する “Systems of Insight” の話。従来の業務システムと最新のクラウド技術を組み合わせてビジネスに必要な “Insight” を得る。

Page 6: 2014 PAPERS BY AUTHOR IEEE - グリッド協議会 A Framework for Dynamic Configuration of Cloud Services Duc-Hung Le, Hong-Linh Truong, Georgiana Copil, Stefan Nastic, Schahram

基調講演(続き) ▪  Scalable HPC System Design and Application ▪  Prof. Yutong Lu (Nantional University of Defense Technology, China) ▪  Tianhe-2 (天河2号) 上でのビッグデータ処理について

▪  Axiomatic, economic, and strategic models of cloud computing ▪  Joe Weinman (IEEE Intercloud Testbed Executive Committee) ▪  クラウド基盤を形式モデル、経済モデル、戦略モデルの観点から分析

▪  Big Data and Cloud Technologies ▪  Dr. Marcel Kunze (Karlsruhe Institute of Technology, Germany) ▪  Smart Data Innovation Lab で行われているビッグデータ処理への取り組みについて。Hadoop の改良やリアルタイム分析等

Page 7: 2014 PAPERS BY AUTHOR IEEE - グリッド協議会 A Framework for Dynamic Configuration of Cloud Services Duc-Hung Le, Hong-Linh Truong, Georgiana Copil, Stefan Nastic, Schahram

セッション: Architecture: Adaptivity ▪  ★Surrogate-Assisted Online Optimisation of Cloud IaaS Configurations Kleopatra Chatziprimou, Kevin Lano, Steffen Zschaler (King’s College London, UK)

▪  SALSA: A Framework for Dynamic Configuration of Cloud Services Duc-Hung Le, Hong-Linh Truong, Georgiana Copil, Stefan Nastic, Schahram Dustdar (Vienna University of Technology, Austria)

▪  ★Elastic Multi-tenant Business Process Based Service Pattern in Cloud Computing Wael Sellami, Hatem Hadj Kacem, Ahmed Hadj Kacem (University of Sfax, Tunisia)

▪  ★A Cluster-based Vehicular Cloud Architecture with Learning-based Resource Management (short paper) Hamid Reza Arkian, Reza Ebrahimi Atani, Atefe Pourkhalili (University of Guilan, Iran)

▪  Automatic Resource Provisioning: a Machine Learning based Proactive approach (short paper) Anshuman Biswas, Shikharesh Majumdar, Biswajit Nandy (Carleton University, Canada), Ali El-Haraki (TELUS, Canada)

…★の付いている論文について紹介

Page 8: 2014 PAPERS BY AUTHOR IEEE - グリッド協議会 A Framework for Dynamic Configuration of Cloud Services Duc-Hung Le, Hong-Linh Truong, Georgiana Copil, Stefan Nastic, Schahram

Surrogate-Assisted Online Optimisation of Cloud IaaS Configurations ▪  目的: ワークロードのスパイクに対する IaaS 設定の最適化をタイムリーに行う

▪  アプローチ: 遺伝的アルゴリズムとして解く(上図)。ただ、毎回シミュレータを動かす訳ではなく、統計的回帰による代理モデル(surrogate-model)によって適応度を計算

▪  結果: Gaussian Processes に基づく代理モデルを使うことで精度(下図)とタイムリーネスをバランスよく両立可能

※ 図は論文からの引用

Page 9: 2014 PAPERS BY AUTHOR IEEE - グリッド協議会 A Framework for Dynamic Configuration of Cloud Services Duc-Hung Le, Hong-Linh Truong, Georgiana Copil, Stefan Nastic, Schahram

Elastic Multi-tenant Business Process Based Service Pattern in Cloud Computing ▪  Multi-tenant な SaaS では、負荷に応じて適切にVMインスタンスを増加させるのは難しい

▪  PaaS と SaaS の間に Cloud Auto-Scaling Middleware (CASM) を入れることで問題を解決 ▪  Auto-scaling のノウハウを “Service pattern” として作成し再利用

▪  Ontology を使って、テナント毎に適切な “Service Pattern” を選択

▪  それに基づいて auto-scale するアルゴリズムを新たに提案

※ 図は論文からの引用

Page 10: 2014 PAPERS BY AUTHOR IEEE - グリッド協議会 A Framework for Dynamic Configuration of Cloud Services Duc-Hung Le, Hong-Linh Truong, Georgiana Copil, Stefan Nastic, Schahram

A Cluster-based Vehicular Cloud Architecture with Learning-based Resource Management ▪  Vehicular Cloud (VC) … Ad-hoc network を介して、CPUやストレージなどのリソースを車車間で互いに融通しあうサービス

▪  Helper … VC の中で、requester の要求に応じてリソース提供を行う車

▪  提案(1): ファジー理論による 柔軟なクラスタリング手法

▪  提案(2): Q-Learning による 適切な helper の選択手法

▪  Omnet++ と SUMO を使い、 既存手法よりも所要時間が 改善されていることを確認

※ 図は論文からの引用

Page 11: 2014 PAPERS BY AUTHOR IEEE - グリッド協議会 A Framework for Dynamic Configuration of Cloud Services Duc-Hung Le, Hong-Linh Truong, Georgiana Copil, Stefan Nastic, Schahram

共著論文の紹介 (in EIC Workshop) ▪  Application-Oriented Bandwidth and Latency Aware Routing with OpenFlow Network Pongsakorn U-chupala, Kohei Ichikawa, Hajimu Iida, Nawawit Kessaraphong, Putchong Uthayopas, Susumu Date, Hirotake Abe, Hiroaki Yamanaka, Eiji Kawai

▪  Performance Characteristics of an SDN-enhanced Job Management System for Cluster Systems with Fat-Tree Interconnect Yasuhiro Watashiba, Susumu Date, Hirotake Abe, Yoshiyuki Kido, Kohei Ichikawa, Hiroaki Yamanaka, Eiji Kawai, Shinji Shimojo, Haruo Takemura

Page 12: 2014 PAPERS BY AUTHOR IEEE - グリッド協議会 A Framework for Dynamic Configuration of Cloud Services Duc-Hung Le, Hong-Linh Truong, Georgiana Copil, Stefan Nastic, Schahram

Application-Oriented Bandwidth and Latency Aware Routing with OpenFlow Network

OpenFlow Network

Controller MonitorApp

REST API

REST API

Monitor by direct measurement

Objective: Align applications’ diverse requirements with different properties of each path in the network and route accordingly to improve application performance and network utilization

Concept of Bandwidth and Latency Aware Network

Structure of Overseer

OpenFlow allows us to control route specifically for each application. Using OpenFlow, we developed Overseer as a reference implementation of BW/LAT aware network

Page 13: 2014 PAPERS BY AUTHOR IEEE - グリッド協議会 A Framework for Dynamic Configuration of Cloud Services Duc-Hung Le, Hong-Linh Truong, Georgiana Copil, Stefan Nastic, Schahram

Performance Characteristics of an SDN-enhanced Job Management System for Cluster Systems with Fat-Tree Interconnect

Average execution time of PING-PONG jobs.

1

10

100

1,000

10,000

1 16 256 4096 65536

Aver

age

exec

utio

n tim

e (s

ec)

Data size (KiB)

OGS/GE

44.0%↓ (64MiB)

Motivation and objective –  For high-performance computing as a cloud service which allows many users to benefit from a large-scale computing system, a new framework for resource management that treats not only the computational resources but also the network resources in the data center is essential.

–  Fat-tree topology has been widely adopted as an interconnect of current cluster system.

SDN-enhanced JMS framework Evaluation result

0.5/1.0 0.8/1.0

0.6/1.0

1.0/1.0

1.0/1.0

J o b

J o b

J o b

J o b

J o b

J o b

J o b

J o b

Network Management Module (NMM)

Database

Brain

Network Control

OpenFlow Controller

(Trema)

Resource Assignment Policy Class Module

User

job job job

Queue JMS (OGS/GE) Computing resource

Management

#!/bin/csh #$ -q QUEUE #$ -pe ompi 2 #$ -l netprio=policy_name mpirun -np 2 ./a.out

Job Script

Administrator

Resource Assignment

Policy 0.9/1.0

Network resource management under OpenFlow

‒  Controlling network flows ‒  Retrieving network resource usage

User can request network resources by requiring the policy name.

Brain decides computational and network resources to allocate to the job according to resource usage and resource assignment policy.

Administrator can create resource assignment policy for computational and network programmablly through a ruby script.