Module 3

4.1 Cloud Computing and Service Models

 The world economy has shifted towards service-oriented industries, with cloud
computing benefiting this sector the most by advancing business computing
with a new paradigm.
 Cloud computing allows developers to rent resources from large, automated
data centers instead of acquiring expensive capital equipment upfront.
 Virtualized cloud platforms, built on large data centers, aim to power the next
generation of data centers by architecting them as virtual resources over
automated hardware, databases, user interfaces, and application environments.
 The fundamental idea is to move desktop computing to a service-oriented
platform using server clusters and large databases in data centers.

4.1.1 Public, Private, and Hybrid Clouds

 Cloud computing evolved from cluster, grid, and utility computing, leveraging
dynamic resources to deliver numerous services to end-users.
 It is a high-throughput computing (HTC) paradigm where infrastructure
services are provided through large data centers or server farms.
 Cloud computing enables users to share access to resources from anywhere, at
any time, through connected devices.

4.1.1.1 Centralized versus Distributed Computing

 Cloud computations are distributed to virtual machines (VMs) in virtual clusters

within data centers, making cloud platforms distributed through virtualization.
 Commercial cloud providers often distribute their platforms geographically for
fault tolerance, reduced response latency, and legal reasons.

4.1.1.2 Public Clouds

 A public cloud is built over the Internet, owned by service providers, and
accessible via subscription to any user who has paid for the service.
 Examples include Google App Engine (GAE), Amazon Web Services (AWS),
and Microsoft Azure.
 They offer application and infrastructure services on a flexible price-per-use
basis.

4.1.1.3 Private Clouds

 A private cloud is built within an organization's intranet, client-owned and

managed, with access limited to the owning clients and their partners.
 They provide a flexible and agile private infrastructure for running service
workloads within administrative domains, aiming for greater customization and
organizational control.

4.1.1.4 Hybrid Clouds

 A hybrid cloud combines both public and private clouds.

 Private clouds can supplement local infrastructure with computing capacity
from an external public cloud to form a hybrid model.
 Hybrid clouds operate with compromises in resource sharing, while public
clouds promote standardization and application flexibility, and private clouds
focus on customization, efficiency, resiliency, security, and privacy.

4.1.1.5 Data-Center Networking Structure

 The core of a cloud is the server cluster, with compute nodes and control nodes
for management and monitoring.
 Clouds are designed to handle fluctuating workloads, demanding dynamic and
variable resources, which private clouds can satisfy if properly managed.
 Data centers differ from supercomputers in scaling requirements and
networking; data centers use IP-based commodity networks optimized for
Internet access, typically with a multilayer structure (Layer 2 for server racks
with switches, Layer 3 for Internet connection with routers).

4.1.1.6 Cloud Development Trends

 While public clouds were dominant in 2010, private clouds are expected to
grow faster due to easier security and trustworthiness within organizations.
 The boundary between public and private clouds may blur, with future clouds
likely being hybrid.
 Cloud computing avoids large data movement during execution, leading to less
Internet traffic and better network utilization.

4.1.2 Cloud Ecosystem and Enabling Technologies

 The cloud computing paradigm follows a pay-as-you-go model, significantly

reducing costs by renting resources instead of purchasing hardware upfront.
 IBM estimated 80% to 95% savings compared to conventional computing,
making it attractive for small businesses.

4.1.2.1 Cloud Design Objectives

 Six key design objectives for cloud computing include:

o Shifting computing from desktops to data centers: Centralizing processing,
storage, and software delivery to data centers over the Internet.
o Service provisioning and cloud economics: Providers offer services via SLAs
with pay-as-you-go pricing, focusing on efficiency in computing, storage, and
power.
o Scalability in performance: Cloud platforms, software, and infrastructure
services must scale with increasing users.
o Data privacy protection: Addressing trust concerns regarding private data in
data centers.
o High quality of cloud services: Standardizing QoS for interoperability among
providers.
o New standards and interfaces: Solving data lock-in by standardizing APIs and
access protocols for portability and flexibility of virtualized applications.

4.1.2.2 Cost Model

 Traditional IT involves fixed capital expenses and variable operational

expenditures, leading to quickly escalating total costs with more users.
 Cloud computing, using a pay-per-use model, has no upfront hardware costs,
only variable operational expenses, significantly reducing costs for users.

4.1.2.3 Cloud Ecosystems

 An ecosystem of providers, users, and technologies has emerged around public

clouds, with growing interest in open-source tools for building IaaS clouds.
 The ecosystem involves four levels: user end (flexible platform demand) , cloud
management (virtualized resources over IaaS) , Virtual Infrastructure (VI)
management (VM allocation over clusters) , and VM management (handling
VMs on host machines).
 Tools like Eucalyptus and Globus Nimbus are open-source for cloud
infrastructure virtualization, accessible via interfaces like Amazon EC2WS.

4.1.2.4 Surge of Private Clouds

 Private clouds leverage existing IT infrastructure and personnel, handling

workloads dynamically and balancing them efficiently within an intranet.
 They provide pre-production testing and enforce data privacy/security policies
more effectively.
 Public clouds offer the major advantage of avoiding capital expenses in IT
investments.

4.1.3 Infrastructure-as-a-Service (IaaS)

 IaaS delivers infrastructure (compute, storage, network) as subscription-based
services in a pay-as-you-go model.
 Amazon Web Services (AWS) is a leader in IaaS, with EC2 providing
virtualized platforms for VMs, S3 for object storage, and EBS for block storage.

4.1.4 Platform-as-a-Service (PaaS) and Software-as-a-Service (SaaS)

 SaaS is often built on top of PaaS, which is built on top of IaaS.

 PaaS provides a platform and environment to create and deploy applications
without the complexity of infrastructure management.
 SaaS provides software applications over the Internet on a subscription basis.

4.1.4.3 Mashup of Cloud Services

 More private clouds are being developed due to a lack of trust in public clouds
for sensitive data, demanding deep customization for enterprise use.

4.2.2 Data-Center Interconnection Networks

 A critical core design of a data center is the interconnection network, which

must meet five requirements: low latency, high bandwidth, low cost, MPI
communication support, and fault tolerance.
 The network topology should support both point-to-point and collective MPI
communication patterns, requiring high bisection bandwidth.

4.2.4 Interconnection of Modular Data Centers

 Container-based data center modules are building blocks for larger data centers.
 The BCube network is used inside server containers, and the MDCube
(Modularized Datacenter Cube) connects multiple BCube containers using high-
speed switches to build large-scale data centers.

4.2.5 Data-Center Management Issues

 Basic requirements for data center management include:

o Making common users happy: Providing quality service for at least 30 years.
o Controlled information flow: Streamlined flow, sustained services, and high
availability (HA).
o Multiuser manageability: Supporting all data center functions like traffic flow
and server maintenance.

4.3.1.3 A Generic Cloud Architecture

 The Internet cloud is envisioned as a massive cluster of servers provisioned on
demand for web services or distributed applications.
 The cloud platform is formed dynamically by provisioning/deprovisioning
servers, software, and database resources, which can be physical machines or
VMs.
 Cloud computing resources are built in data centers typically owned and
operated by third-party providers, where software becomes a service.

4.3.2.2 Quality of Service Factors

 Key QoS parameters for cloud services include time, cost, reliability, and
trust/security, which are not static and may change over time.

4.3.3 Data Replication and Disaster Recovery

 Cloning VMs offers an effective solution for disaster recovery, making a clone
VM on a remote server for every running VM on a local server.
 A cloud control center should activate a suspended clone VM in case of failure
of the original VM, taking snapshots for live migration with minimal downtime.

4.3.4 Architectural Design Challenges

 Challenge 1: Service Availability and Data Lock-in Problem:

o Single company management can lead to single points of failure; using multiple
cloud providers enhances HA.
o Distributed denial of service (DDoS) attacks are an availability obstacle.
o Proprietary APIs cause data lock-in; standardization of APIs is needed for
portability across providers and to enable "surge computing".
 Challenge 2: Data Privacy and Security Concerns:
o Public networks expose systems to attacks. Encrypted storage, virtual LANs,
and firewalls can mitigate some risks.
o Newer attacks in cloud environments include hypervisor malware, guest
hopping/hijacking, and VM rootkits.

4.4.2 Google App Engine (GAE)

 GAE is a PaaS that supports many web applications, providing a storage service
for application-specific data in the Google infrastructure.
 It offers persistent data storage with query, sorting, and transaction capabilities,
similar to traditional databases.
 GAE provides Google-specific services like Gmail account service for user
authentication and email sending, eliminating the need for custom user
management.
 4.4.3 Amazon Web Services (AWS)

 Amazon is a leader in public cloud services, applying the IaaS model.

 EC2 (Elastic Compute Cloud): Provides virtualized platforms to host VMs for
cloud applications.
 S3 (Simple Storage Service): Offers object-oriented storage services.
 EBS (Elastic Block Service): Provides block storage for traditional
applications.
 AWS also supports auto-scaling (automatically adjusting VM instance capacity)
and elastic load balancing.

4.5.1.1 Cloud Service Tasks and Trends

 Cloud services are introduced in five layers, with SaaS applications at the top
(e.g., CRM for business), followed by PaaS, IaaS, collocation services (multiple
providers), and network cloud services.
 Cloud mashup is often practiced in vertical cloud applications.

4.5.1.2 Software Stack for Cloud Computing

 The overall software stacks for cloud computing clusters are built from scratch
to meet rigorous goals like high throughput, HA, and fault tolerance.
 The software stack structure can be viewed in layers, with even the operating
system potentially modified for cloud data processing