Cloud Computing A Trend Taking Shape Yangfan Zhou

Cloud Computing A Trend Taking Shape Yangfan Zhou Dept. Computer Science and Engineering The Chinese University of Hong Kong Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 1

Contents Drive of cloud computing Nature of cloud computing Cloud computing industry Trustworthy cloud computing Conclusions Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 2

Contents Drive of cloud computing Nature of cloud computing Cloud computing industry Trustworthy cloud computing Conclusions Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 3

What the buzzword is about You may ask: Is cloud computing • created by marketing people? ? A new name of an old and awful technology? ? • a new technology focusing on higher-speed computing? ? • a new technology that can let you use more storage capacity? ? • a new concept that can help programmers? maintainers? Let’s see what it is via its economic drive Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 4

Hosting computing systems Server CSE co. ltd. Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 5

Hosting computing systems CSE co. ltd. Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 6

Hosting computing systems CSE co. ltd. Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 7

Hosting computing systems Peak hour Most of the time Waste of resource = waste of $ CSE co. ltd. Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 8

Hosting computing systems Waste of resource = waste of $ • You pay for the servers no matter whether you are using them or not – The money you spend to buy the hardware – The salary for the maintainers – The electricity (including that for air conditioning) – Other maintenance costs, e. g. , repairing, upgrading, renting a room for the servers. CSE co. ltd. Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 9

Hosting computing systems How I I can provide a So that wish this cloud can really host the scalable, on-demand computing systems for service. me Server CSE co. ltd. Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 10

When there is a dream, there is a market I don’t have to care what it is. I can just host my services here. • A lot of companies said: Sure!!! We can produce such a cloud for you – Microsoft, IBM, Google, Amazon, HP, Yahoo!, Intel. . • A lot of people do have done a lot of hard marketing work – Many more IT companies join and make the same promise before they really know what the promise means technologically – Try to define “cloud computing” – Try to register “cloud computing” as a trademark – Found many startups Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 11

When there is a dream, there is a market A recent reaction in the capital market PE=Price per Share/Earning per Share Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 12

When there is a dream, there is a market Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 13

Contents Drive of cloud computing Nature of cloud computing Cloud computing industry Trustworthy cloud computing Conclusions Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 14

Cloud computing preliminaries Computing System User software, API Application OS, middleware hardware Platform Infrastructure Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 15

Cloud computing preliminaries Three cloud computing stacks A cloud user’s own application and Platform OS Saa. S: Software as a service Infrastructure Paa. S: Platform as a service Provided by the cloud computing provider Iaa. S: Infrastructure as a service Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 16

Pay as you go The core of cloud computing is computing/storage outsourcing Out-sourcing is a great idea!!! vs Pay as you go!! Pay for exactly what you’ve used Cost Down!!! Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 17

Outsourcing Electronics industry Service vendor Brand vendor Cloud computing provider Manufacturing outsourcing Different manufacturing requirements in low season and high season # of product lines # of workers OEM: original equipment manufacturer ODM: original design manufacturer Services for end users Computing and storage outsourcing Different service requirements in different times # of servers # of maintainers Cost down!!! Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong OSP: original service provider Cost down!!! 18

Cloud computing v. s. Grid computing Cloud computing is not just a new term for an old idea • Grid computing – Focus on let more than one computer coordinate to solve a problem together – Often include heterogeneous environments, i. e. , computers of different capacity, configuration, and even different OS. – Business Model: project-oriented • Cloud computing – An application doesn't access resources directly. It accesses them through a service. Usually the service has access to a large amount of physical resources, and can dynamically allocate them on demand. – Business Model: pay on a consumption basis A cloud can use grid computing technique. But a grid is not necessarily a cloud or part of a cloud. Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 19

Contents Drive of cloud computing Nature of cloud computing Cloud computing industry Trustworthy cloud computing Conclusions Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 20

Cloud computing industry Representative Providers • • Amazon Elastic Compute Cloud (EC 2) Google App Engine Microsoft's Windows Azure Platform Other small startups: Heroku & Engine Yard Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 21

Amazon Elastic Compute Cloud (EC 2) • Xen-based virtual computing environment where a user can run Linux-based applications • A user can boot an Amazon Machine Image to create a virtual machine instance containing any software desired • Iaa. S: A user can control nearly the entire software stack, from the kernel upwards. It provides low level of virtualization – raw CPU cycles, block-device storage, IP-level connectivity • Provided together with Simple Storage Services (S 3) – data storage service • A user can increase or decrease capacity within minutes • Amazon charges $0. 084/hour ($61/month) for the smallest virtual machine and around $0. 10 per gigabyte of data transfer Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 22

Google App Engine • Application domain-specific platforms. Not suitable for general-purpose computing • Allow a user to run Web applications in Python or Java. Host applications in Google-managed data centers • Enforcing an application structure of clean separation between a stateless computation tier and a stateful storage tier • Support APIs for the Google Datastore, Google Accounts, URL fetch, image manipulation, and email services • Automatically scale in response to load increases and decreases, and users are charged by the cycles used • Free up to a certain level of used resources – 500 MB of storage and about 5 million page views Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 23

Microsoft's Windows Azure Platform • Use Windows Azure Hypervisor as the infrastructure • Use the. NET framework as the application container and applications are compiled to the Common Language Runtime, a language-independent managed environment • Paa. S: Can be considered as something intermediate between application frameworks like Google App Engine and hardware virtual environments like Amazon EC 2 • Support general-purpose computing, rather than a single category of application • Users can choose language, but cannot control the underlying operating system or runtime • The libraries provide a degree of automatic network configuration and failover/scalability, but require the developer to declare some application properties Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 24

Other small startups: Heroku & Engine Yard • Heroku – Based on Ruby on Rails – Dyno: a dyno is a single process running ruby code on a server in Heroku grid – Application codes run inside the dyno grid, occupying as many slots as needed. New dynos for an application can be started in under 2 seconds for most applications. – Dynos are launched in an environment containing your app's database and cache information – Charge on a capacity basis. Free for the entry level • Engine Yard – Also based on Ruby on Rails – Similar Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 25

Contents Drive of cloud computing Nature of cloud computing Cloud computing industry Trustworthy cloud computing Conclusions Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 26

Cloud computing challenges and opportunities Why people will trust their data/computing to the cloud and accept the idea of computing/storage outsourcing? ? Challenges and opportunities of our interests • How to provide a scalable service, adaptive to the diversity of end-users (# of end users, locations of end users) – Qo. S provisioning • How to provide a secure service – Privacy preserving – Secure access guaranteeing – Resilient to attacks • How to provide a reliable service – Fault tolerance Towards Trustworthy Cloud Computing – New test methodologies Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 27

Cloud computing challenges and opportunities Topic 1: Qo. S Provisioning How to provide a scalable service, adaptive to the diversity of end-users Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 28

Cloud computing challenges and opportunities • Preliminaries of service composition – A service can be composed by many components – Components may locate world-wide – There are many components with the same functionalities components ` Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong A cloud service 29

Cloud computing challenges and opportunities Potential research problems Qo. S driven recommendation for cloud service composition • Observations – The Qo. S properties of the same cloud service component are different in the users’ perspective • Users are located world wide • The network connections of different users are different – Software quality metrics no longer depends only on the software per se Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 30

Cloud computing challenges and opportunities Potential research problems • Problem 1: Qo. S driven recommendation for cloud service composition: How to recommend a set of components to form a service based on a cloud user’s networking feature – Heuristic: recommend based on IP address – Recommend based on the similarity – Mapping our Web services research to this domain Component 1 Component 2 Component 3 Component 4 User 1 Value 12 Value 13 Value 14 User 2 Value 21 Value 22 Value 23 Value 24 User 3 Value 31 Value 32 Value 33 Value 34 New User Prediction Value 43 Value 44 Prediction Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 31

Cloud computing challenges and opportunities Potential research problems • Problem 2: How many resources/components that should be proactively reserved to guarantee Qo. S – Based on a prediction of the service usage – We should look into how resources are virtualized Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 32

Cloud computing challenges and opportunities Potential research problems • Problem 3: Real-world measurement on the existing clouds – Amazon EC 2, Google APP Engine – Elastic? – Qo. S all over the world? Commercial Cloud Planet. Lab Node Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 33

Cloud computing challenges and opportunities Topic 2: Fault tolerance and reliability How to provide a reliable service given the fact no components are perfectly created, and the Internet is unpredictable Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 34

Cloud computing challenges and opportunities Potential research problem • Problem 1: Reliability prediction of cloud service composition, i. e. , how to predict the reliability quality of a service that is composed by a set of components – Previous method: measurement • Cannot know before deployment • Cannot know if we cannot do tremendous tests • Hard to know if we don’t know how components are coded – Predict also based on the similarity Component 1 Component 2 Component 3 Component 4 User 1 Value 12 Value 13 Value 14 User 2 Value 21 Value 22 Value 23 Value 24 User 3 Value 31 Value 32 Value 33 Value 34 New User Prediction Value 43 Value 44 Prediction – Reliability can be predicted in the service design phase based on its composition specific Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 35

Cloud computing challenges and opportunities Potential research problem • Problem 2: How to tolerate faults in stateful cloud applications. – Long term business process – Cloud computing: Redundancy in nature – Solutions • Recovery – Check pointing and rollback • Standby redundancy – State transfer • Parallel redundancy – State synchronization. Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 36

Cloud computing challenges and opportunities Other potential research problems • How to exploit the redundancy in cloud computing • Cloud failure detection • Component dependency analysis • Error propagation analysis Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 37

Cloud computing challenges and opportunities Topic 3: Security How to provide a secure service Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 38

Cloud computing challenges and opportunities Potential research problems • Problem 1: Identify critical cloud nodes – Dynamic service composition Complex component invoking relations – Reliability and attack-resilience features of different components may have different impacts on the whole cloud – Identify critical components • Purpose: To reduce the vulnerability • Approach: based on the service composition relations (Ranking) Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 39

Cloud computing challenges and opportunities Potential research problems • Problem 2: Attack resilient – Subproblem 1: Do. S masking • Utilize the redundant nature of components • Intelligent watchdog technique • Load balance technique – Subproblem 2: Anti-Masquerade attacks • A malicious node will fake a large number of pseudonymous components, which can be used to gain a biased trust in service composition • Design a reputation mechanism • Use similar techniques as those for anti-spam in searching engines Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 40

Cloud computing challenges and opportunities Potential research problems • Problem 3: New access control paradigm – Purpose: • Anti-freeriding • Guarantee data security – Challenge: How to share the user credentials among different cloud nodes/components efficiently • A user may login at a node • And invoke a component at another node • Access control paradigm for stateful services Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 41

Cloud computing challenges and opportunities Potential research problem? • Privacy preserving – – – There already papers in this area. Can not be a good area to publish May still be a big deal, because … academic papers Several top conference private We may have to provide papers data to the cloud because we need it to compute them Email servers You put your emails there without worrying about leaking your privacy? Why we trust the IT guys in our department more than a cloud computing company far away We have laws and commercial moral. Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 42

Cloud computing challenges and opportunities Topic 4: Testing How to test cloud computing system Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 43

Cloud computing challenges and opportunities Potential research problems • Problem 1: Embracing user-supplied software testing – Test-as-you-use paradigm is possible Software is running at servers, runtime data can be collected easily Software is running at PCs, runtime data can only be collected by user report – Tremendous runtime data can greatly facilitate testing Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 44

Cloud computing challenges and opportunities Potential research problems • Problem 2: Anomaly detection (for locating potential faults) – Test-as-you-use paradigm is possible – There are tremendous runtime data – Can these data be utilized for locating potential faults • How to locate anomaly • How to locate the root cause of the anomaly Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 45

Cloud computing challenges and opportunities Topic 5: System How to build a cloud computing system Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 46

Cloud computing challenges and opportunities Potential research problems Problem 1: Building a cloud • An organization provides its PCs as nodes in the cloud • Users can use the cloud • Challenges – Many engineering efforts • Which language(s) it supports • The framework: Iaa. S or Paa. S • More including coding – – – Accounting system Reliability system Qo. S provisioning Attack-resilient And many others Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 47

Cloud computing challenges and opportunities Potential research problems Problem 2: Building an open cloud • Everyone can provide his/her PC as a node in the cloud • And everybody can use the cloud • Challenges – Many engineering efforts • Which language(s) it supports • The framework: Iaa. S or Paa. S • More including coding – A reputation mechanism to stimulate contributions – A reputation mechanism for anti-freeriding – Attack-resilient Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 48

Current proposal outline • Task 1: Achieving trustworthy cloud computing in architectural design – Ranking-based identification of critical cloud nodes – Qo. S driven recommendation for cloud service composition – Development of an open cloud computing architecture • Task 2: Enhancing the security and reliability for cloud computing paradigm – A new access control paradigm – Tolerating faults in cloud computing – Designing a reputation mechanism for an open cloud • Task 3: Formulating a new testing methodology for trustworthy cloud computing – Embracing user-supplied software testing in cloud computing paradigm – Unveiling faults via symptom mining • Task 4: Implementing an open-source cloud computing framework – Test-bed construction – Large scale experiments Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 49

Conclusions 1. Cloud computing include Iaa. S, Paa. S, Saa. S 2. We identify several research areas 1. 2. 3. 4. 5. Qo. S provisioning Fault tolerance and reliability Security Testing Cloud system implementations 3. Cloud service composition is a critical feature, enriching Qo. S provisioning, fault tolerance, and reliability issues 4. 4. Redundancy is an important nature of cloud computing systems, enabling fault tolerance, attack resilience 5. Application runtime data are available in the cloud, facilitating reliability prediction, Qo. S provisioning, and testing Dept. of Computer Sci. & Eng. , The Chinese University of Hong Kong 50