IaaS Cloud Architectures from Virtualized Data Centers to Federated Cloud Infrastructure

IaaS Cloud Architectures:
Virtualized Data Centers to
Federated Cloud Infrastructures
Dr. Sanjay P. Ahuja, Ph.D.
2010-14 FIS Distinguished Professor of Computer
Science
School of Computing, UNF

Introduction
 Virtualization plays an important role as an enabling technology for
datacenter implementation by abstracting compute, network, and
storage service platforms from the underlying physical hardware.
 Because secure, efficient, and scalable management of these
virtualized infrastructures is essential to guarantee optimal
datacenter operation, the virtual infrastructure manager is a key
component of the datacenter architecture.
 The virtual infrastructure manager, also called the cloud operating
system (cloud OS), orchestrates the deployment of virtual
resources and manages the physical and virtual infrastructures.
 Management of the datacenter as a cloud makes it possible to
complement the local infrastructure with remote resources from
other federated datacenters or commercial clouds.

Introduction (contd.)
 This cloud vision of the datacenter provides several advantages
including server consolidation to reduce hardware and power
requirements, on-the-fly resizing of the physical infrastructure,
service workload balance among physical resources to improve
efficiency and utilization, server replication to support fault tolerance
and high availability capabilities, and dynamic partitioning of
physical infrastructure to execute and isolate different services and
workloads.
 Turning this vision into reality requires developing an open and
flexible cloud architecture reference model that addresses the
requirements of business use cases from IT companies. The model
must provide interoperability and portability to prevent vendor lock-
in, scalability, and standardization.

Challenge in Cloud Computing
 An open challenge in cloud computing is cloud federation, which
involves different architectures and levels of coupling among
federated cloud instances.
 The cloud architecture reference model must include support for the
deployment of different federation scenarios so that cloud providers
and IT companies can use external resources as well as share their
internal resources.

Cloud Infrastructure Anatomy
 The key component of an IaaS cloud architecture is the cloud OS
(see figure in next slide), which manages the physical and virtual
infrastructures and controls the provisioning of virtual resources
according to the needs of the user services.
 A cloud OS’s role is to efficiently manage datacenter resources to
deliver a flexible, secure, and isolated multitenant execution
environment for user services that abstracts the underlying physical
infrastructure and offers different interfaces and APIs for interacting
with the cloud.
 While local users and administrators can interact with the cloud
using local interfaces and administrative tools that offer rich
functionality for managing, controlling, and monitoring the virtual and
physical infrastructure, remote cloud users employ public cloud
interfaces that usually provide more limited functionality.

The Cloud OS
As a key component in a modern datacenter, the cloud
operating system is responsible for:
1. managing the physical and virtual infrastructure,
2. orchestrating and commanding service provisioning and
deployment,
3. providing federation capabilities for accessing and
deploying virtual resources in remote cloud
infrastructures

Infrastructure and Cloud Drivers
 To provide an abstraction of the underlying infrastructure, technology, the
cloud OS can use adapters or drivers to interact with a variety of
virtualization technologies. These include hypervisor, network, storage, and
information drivers.
 The core cloud OS components, including the virtual machine (VM)
manager, network manager, storage manager, and information manager,
rely on these infrastructure drivers to deploy, manage, and monitor the
virtualized infrastructures.
 In addition to the infrastructure drivers, the cloud OS can include different
cloud drivers to enable access to remote providers.
 OpenNebula (http://opennebula.org) is an example of an open cloud OS
platform focused on datacenter virtualization that fits with the architecture
proposed in Figure 1. Other open cloud managers, such as OpenStack
(http://openstack.org) and Eucalyptus (www.eucalyptus.com), primarily
focus on public cloud features.

Virtual Machine Manager
 A cloud OS defines the VM as the basic execution unit and the
virtualized services (group of VMs for executing a multitier service)
as the basic management entity.
 This concept helps create scalable applications because the user
can either add VMs as needed (horizontal scaling) or resize a VM (if
supported by the underlying hypervisor technology) to satisfy a VM
workload increase (vertical scaling).
 Individual multitier applications are isolated from each other, but
individual VMs in the same applications are not, as they all can
share a communication network and services when needed.
 A VM consists of a set of parameters and attributes, including the
OS kernel, VM image, memory and CPU capacity, network
interfaces etc.

Virtual Machine Manager (contd.)
 The VM manager is responsible for managing a VM’s entire life
cycle and performing different VM actions—deploy, migrate,
suspend, resume, shut down—according to user commands or
scheduling strategies.
 To perform these actions, the VM manager relies on the hypervisor
drivers, which expose the basic functionality of underlying
hypervisors such as Xen, KVM, and VMware to avoid limiting the
cloud OS to a specific virtualization technology.
 The VM manager is also responsible for preserving the service-level
agreements contracted with the users, which are usually expressed
in terms of VM availability in infrastructure clouds. To guarantee this
availability, the VM manager should include different mechanisms
for detecting VM crashes and automatically restarting the VM in
case of failure.

Network Manager
 The deployment of services in a cloud involves not only the provision of
VMs but also the instantiation of communication networks to interconnect
the different service components and to make the service reachable for
external users, if needed.
 The network manager should be able to manage private networks to
interconnect both the service’s internal components and public IP address
pools and connect the front-end service components to the Internet.
 The network manager uses the network drivers to provision virtual networks
over the physical network infrastructure. It needs to ensure traffic isolation
between virtual networks.

Storage Manager
 The storage manager’s main function is to provide storage services
and final-user virtual storage systems as a commodity.
 The storage system must be scalable so that it can grow
dynamically according to service needs; highly available and
reliable, to avoid data access disruption in data access in case of
failure; high-performance, to support strong demands of data-
intensive workloads; and easy to manage, abstracting users from
the underlying physical storage’s complexity.
 To reach these goals, the storage manager relies on the existing
storage drivers, which introduce a layer of abstraction between
users or services and physical storage and enable the creation of a
storage resource pool where storage devices appear as one,
allowing data to be moved freely among devices.

Image Manager
 Image managers must handle a huge amount of VM images
belonging to different users, with different operating systems and
software configurations.
 Thus, the cloud OS must have the appropriate tools to manage these
images efficiently and securely, as well as having additional functionality for
administering image repositories.
 A set of attributes defines the VM image, including the image’s name, a
description of its contents, the type of image—public, private, or shared—
the image owner, and the image’s location within the repository.
 Basic image functionality should include tools for creating a new image in a
repository, deleting an image, cloning an image from an existing one,
adding or changing an image attribute, sharing an image with other users,
publishing an image for public use, or listing the images available in the
repository.

Information Manager
 The information manager is responsible for monitoring and gathering
information about the state of VMs, physical servers, and other
components of virtual and physical infrastructures such as network
devices and storage systems. This monitoring function is essential
to ensure that all these components are performing optimally.
 The information manager uses the information drivers to collect monitoring
information from virtual and physical resources. At the physical server level,
the administrator can install different specialized tools for monitoring
purposes such as Nagios (www.nagios.com) and Ganglia
(http://ganglia.sourceforge.net).
 Monitoring at the VM level relies on the information provided by hypervisors,
which can be very limited and might differ from one hypervisor to another.

Authentication and Authorization
 Clouds must incorporate mechanisms to authenticate users and
administrators and to provide them with access only to authorized
resources.
 User authentication verifies and confirms the identity of users who try to
access cloud resources. This function can be implemented using different
methods, such as simple password verification mechanisms via LDAP;
trusted authentication mechanisms based on public keys, X.509 certificates
or Kerberos.
 Authorization policies control and manage user privileges and permissions
to access different cloud resources, such as VMs, networks, or storage
systems. Access control can be implemented using role-based
mechanisms, where a role defines a group of permissions to perform certain
operations over specific cloud resources and users can be assigned
particular roles.
 Quota mechanisms can be used to limit the amount of resources—CPU,
memory, network bandwidth, or disk—a specific user can access.

Accounting and Auditing
 Accounting’s objective is to obtain and record resource usage
information of deployed services. This function relies on the
information manager to monitor resources and collect usage
information from metric measurements. Accounting is essential to
implementing the mechanisms that produce billing information.
 Auditing provides information about activity in cloud resources,
indicating who accessed cloud resources, when they gained access,
and what operations they performed. This information is useful to
improve cloud security and protect it from threats such as
unauthorized access, abusive use of resources, and other forms of
intrusion.

Federation Manager
 The federation manager enables access to remote cloud
infrastructures, which can be either partner infrastructures governed
by a similar cloud OS entity or public cloud providers.
 The federation manager should provide basic mechanisms for
deployment, runtime management, and termination of virtual
resources in remote clouds; remote resource monitoring; user
authentication in remote cloud instances; access control
management and remote resource permission; and tools for image
building on different clouds with different image formats.
 The federation manager’s design could differ depending on the
supported types of federation—for example, cloud aggregation,
bursting, or brokering—and levels of coupling and interoperability.
The cloud OS must implement the federation manager as an internal
component to support federation architectures at the infrastructure
level.

Scheduler
 There are two levels of scheduling within a cloud infrastructure: at
the physical host level, managed by the hypervisor scheduler, which
is responsible for deciding when VMs can obtain system
resources—such as physical CPU or memory—and which physical
CPUs are assigned to each VM; and at the cloud level, managed by
the cloud OS scheduler, which is responsible for deciding the
particular physical server where each VM is deployed.
 The cloud OS scheduler’s main function is to decide the initial
placement of each VM following specific criteria.
 In a federated environment, the scheduler could decide to deploy
the VM in a remote cloud when insufficient resources are available
in the local infrastructure.

Scheduler (contd.)
 The scheduler could also provide dynamic optimization capabilities,
enabling the dynamic reallocation (migration) of VMs from one
physical resource to another to meet specific optimization criteria.
 The table in the next slide lists different scheduling policies, based
on varying optimization criteria, to guide both initial placement and
dynamic reallocation actions.
 The user can also specify the constraints that can restrict scheduler
decisions such as, for example, hardware (amount of CPU, memory,
and so on), platform (type of hypervisor, OS, and so on), affinity (two
or more VMs that need to be deployed in the same physical server
or the same physical cluster), location (geographical restrictions), or
service-level agreement constraints (guaranteed CPU capacity or
high operational reliability).

Optimization Criteria for Allocation and Reallocation
Policies of the Scheduler

Scheduler (contd.)
 The cloud OS invokes the scheduler every time a new VM is waiting
to be deployed as well as periodically to optimize the entire virtual
infrastructure, reallocating VMs if necessary to meet the established
optimization criteria.
 The scheduler interacts with the VM manager to deploy or allocate
the VM in the selected server or with the federation manager to
deploy VMs in remote clouds.

Administrative Tools
 The cloud OS must provide different tools and interfaces (command-
line or GUI) for users and administrators to perform various tasks.
 For privileged administration, the system should include both user
administration tools (to create, modify, or delete users and manage
user authorization and access control policies) and physical
infrastructure management tools (to boot or shut down physical
servers, monitor physical infrastructure, and so on).
 Unprivileged users should also be provided with tools to manage
their own infrastructure—for example, VM management tools (to
deploy, shut down, suspend, restore, or monitor a VM), virtual
network management tools (to create or delete virtual networks),
virtual storage management tools (to create, delete, or attach a
virtual disk), and image management tools (to create, clone, or
delete images).

Service Manager
 The cloud OS should be able to manage and support virtualized multitier
services. A multitier service can comprise several component/tiers with
some intrinsic dependencies among them. These services can be deployed
as a group of interconnected VMs in the cloud with specific deployment
dependencies and, optionally, some location, affinity, and elasticity
requirements.
 The service manager’s admission control function entails deciding whether
to accept or reject a service, depending on the service requirements and
resource availability in the cloud.
 Once it accepts a service, the service manager is responsible for managing
its life cycle, which can involve several actions, including deploying,
suspending, resuming, or canceling the service.
 To deploy a new service, the service manager interacts with the scheduler
to decide the best placement for the various VMs that comprise the service,
according to the selected optimization criteria and service constraints.

Service Manager (contd.)
 Another service manager function is the management of service
elasticity. The service manager can incorporate different
mechanisms for service auto-scaling based on elasticity rules, which
trigger the deployment of new instances (horizontal scaling) or by
resizing existing instances (vertical scaling) when user-specified
service metrics exceed certain thresholds.
 Independent of the service manager, users are always allowed to
employ the interfaces provided by the administrative tools or the
cloud interface to deploy, resize, migrate, or shut down their
individual VMs.

Cloud Interfaces
 In the current cloud ecosystem, most cloud products and providers
offer their own APIs, such as Amazon EC2 or VMware’s vCloud.
Although some of these APIs are becoming de facto standards, this
heterogeneity makes it difficult to achieve interoperability and
portability across clouds.
 Several standards bodies are addressing interoperability and
portability issues surrounding cloud infrastructures. Example, OGF
OCCI (http://occi-wg.org), OVF (http://dmtf.org/standards/cloud).

Cloud Federation
 Cloud federation, which enables cloud providers and IT companies to
collaborate and share their resources, is associated with many portability
and interoperability issues.
 Cloud developers and researchers have proposed or implemented
numerous federation architectures, including cloud bursting, brokering,
aggregation, and multitier.
 These architectures can be classified according to the level of coupling or
interoperation among the cloud instances involved, ranging from loosely
coupled (with no or little interoperability among cloud instances) to tightly
coupled (with full interoperability among cloud instances).

Loosely Coupled Federation
 This scenario is formed by independent cloud instances—for
example, a private cloud complementing its infrastructure with
resources from an external commercial cloud—with limited
interoperation between them.
 A cloud instance has little or no control over remote resources (for
example, decisions about VM placement are not allowed),
monitoring information is limited (for example, only CPU, memory, or
disk consumption of each VM is reported), and there is no support
for advanced features such as cross-site networks or VM migration.

Partially Coupled Federation
 This scenario typically consists of various partner clouds that
establish a contract or framework agreement stating the terms and
conditions under which one partner cloud can use resources from
another.
 This contract can enable a certain level of control over remote
resources (for example, allowing the definition of affinity rules to
force two or more remote VMs to be placed in the same physical
cluster); can agree to the interchange of more detailed monitoring
information (for example, providing information about the host where
the VM is located, energy consumption, and so on); and can enable
some advanced networking features among partner clouds (for
example, the creation of virtual networks across site boundaries).

Tightly Coupled Federation
 This scenario usually includes clouds belonging to the same
organization and is normally governed by the same cloud OS type.
 In this scenario, a cloud instance can have advanced control over
remote resources—for example, allowing decisions about the exact
placement of a remote VM—and can access all the monitoring
information available about remote resources.
 In addition, it can allow other advanced features, including the
creation of cross-site networks, cross-site migration of VMs,
implementation of high availability techniques among remote cloud
instances, and creation of virtual storage systems across site
boundaries.

Cloud Federation Architectures

Bursting (Hybrid) Architecture
 Cloud bursting or hybrid architecture combines the existing on-
premise infrastructure (usually a private cloud) with remote
resources from one or more public clouds to provide extra capacity
to satisfy peak demand periods.
 Because the local cloud OS has no advanced control over the virtual
resources deployed in external clouds beyond the basic operations
the providers allow, this architecture is loosely coupled. Most
existing open cloud managers support the hybrid cloud architecture
and is used in infrastructures such as StratusLab
(http://stratuslab.eu).

Broker Architecture
 The central component of the broker architecture is a broker that
serves various users and has access to several public cloud
infrastructures. A simple broker should be able to deploy virtual
resources in the cloud as selected by the user.
 An advanced broker offering service management capabilities could
make scheduling decisions based on optimization criteria such as
cost, performance, or energy consumption to automatically deploy
virtual user service in the most suitable cloud, or it could even
distribute the service components across multiple clouds. This
architecture is also loosely coupled since public clouds typically do
not allow advanced control over the deployed virtual resources.
 Brokering is the most common federation scenario. Examples
include BonFIRE (www.bonfire-project.eu), Open Cirrus, and

Aggregated Architecture
 Cloud aggregation consists of two or more partner clouds that
interoperate to aggregate their resources and provide users with a
larger virtual infrastructure. This architecture is usually partially
coupled, since partners could be provided with some kind of
advanced control over remote resources, depending on the terms
and conditions of contracts with other partners.
 These partner clouds usually have a higher coupling level when they
belong to the same corporation than when they are owned by
different companies that agree to cooperate and aggregate their
resources. The Reservoir federated infrastructure is an example of
an aggregated cloud architecture.

Multitier Architecture
 The multitier architecture consists of two or more cloud sites, each
running its own cloud OS and usually belonging to the same
corporation, that are managed by a third cloud OS instance following
a hierarchical arrangement.
 This upper cloud OS instance has full control over resources in
different cloud sites—a tightly coupled scenario—and it exposes the
resources available in the different cloud sites as if they were
located in a single cloud.
 This architecture is beneficial for corporations with geographically
distributed cloud infrastructures because it provides uniform access.
It is also useful for implementing advanced management features
such as high availability, load balancing, and fault tolerance.

Reference and Acknowledgement
The contribution of the following paper is duly acknowledged for this
presentation:
 Moreno-Vozmediano, R., Montero, R., and Llorente, I., "IaaS Cloud
Architecture: From Virtualized Data Centers to Federated Cloud
Infrastructures", IEEE Computer Vol. 45 (12), Dec. 2012, pp. 65-72.

IaaS Cloud Architectures from Virtualized Data Centers to Federated Cloud Infrastructure

More Related Content

Similar to IaaS Cloud Architectures from Virtualized Data Centers to Federated Cloud Infrastructure (20)

Recently uploaded (20)

IaaS Cloud Architectures from Virtualized Data Centers to Federated Cloud Infrastructure