Cloud Computing

Assignment 1-2

1. Explain the concept of scalable computing. How has

internet computing evolved over time, and what role
does high-performance computing (HPC) and high-
throughput computing (HTC) play in its development?
Answer:
Scalable computing refers to the ability of a computing system to handle increasing work-
loads by either scaling up (vertical scaling) or scaling out (horizontal scaling). Vertical
Scaling (Scale-Up): Adding more power (CPU, RAM, storage) to an existing machine.
Horizontal Scaling (Scale-Out): Adding more machines (servers/nodes) to distribute
the workload.
This concept ensures efficient resource utilization, cost-effectiveness, and adaptability
to growing demands in computing environments.

Evolution of Internet Computing

Internet computing has evolved through multiple phases:
1. Mainframe Computing (1950s–1970s):

• Large centralized computers with multiple terminals

• Limited scalability due to hardware constraints.

2. Client-Server Computing (1980s–1990s):

• Introduction of personal computers connected to powerful centralized servers.

• More scalable but still dependent on hardware limits.

3. Grid and Cluster Computing (Late 1990s–2000s):

• Cluster Computing: Multiple computers work as a single system.

• Grid Computing: Distributes tasks across geographically dispersed resources.
• Used in scientific research and data-intensive applications.

4. Cloud Computing (2000s–Present):

• On-demand resources (IaaS, PaaS, SaaS) with scalable architectures.

• Global accessibility with providers like AWS, Azure, Google Cloud.
• Enabled scalable computing with virtual machines, containers, and microser-
vices.

5. Edge Fog Computing (2015–Present):

• Processing data closer to the source (IoT, AI-driven applications).

• Reduces latency and optimizes bandwidth usage.

Govt Engineerig College MH Halli 1 Department of CSE

 Cloud Computing

Role of HPC and HTC in Scalable Computing

1. High-Performance Computing (HPC):

• Uses supercomputers and clusters to process complex simulations and compu-

tations.
• Involves parallel processing to solve scientific, engineering, and AI problems.
• Example: Weather modeling, genomics, AI training, CFD simulations.

2. High-Throughput Computing (HTC):

• Focuses on maximizing computational capacity over time.

• Often used in cloud and grid computing to process large numbers of smaller
tasks.
• Example: Data analytics, bio-informatics, financial modeling, rendering.

article array booktabs

Evolution Phase Key Features Role of HPC/HTC

Mainframe (1950s) Centralized, slow Minimal HPC usage
Client-Server Distributed, LAN Basic HPC clusters
Grid Computing Global resource Early HPC usage
Cloud Computing On-demand, virtual HTC for workloads
Edge Computing Low-latency, AI HPC + AI-driven

Table 1: Evolution of Computing and Role of HPC/HTC

2.Discuss the major trends in scalable computing. How

have emerging paradigm influenced scalable comput-
ing over the internet?
Answer:
Major Trends in Scalable Computing and the Influence of Emerging Paradigms Scalable
computing has evolved significantly due to advances in hardware, software, and network-
ing technologies. This evolution has been influenced by various predictable trends, such
as Moore’s Law and Gilder’s Law, which have shaped the way distributed and parallel
computing systems are designed and implemented.

Major Trends in Scalable Computing

1. Degrees of Parallelism in Computing

• Scalable computing has witnessed continuous improvements in parallelism,

leading to different levels of parallel processing.

Govt Engineerig College MH Halli 2 Department of CSE

Cloud Computing

• Bit-Level Parallelism (BLP): Improved processing by transitioning from 4-bit

to 64-bit microprocessors.
• Instruction-Level Parallelism (ILP): Enabled multiple instructions to execute
simultaneously through techniques like pipelining, superscalar computing, VLIW
(Very Long Instruction Word), and multithreading.
• Data-Level Parallelism (DLP): Utilized SIMD (Single Instruction, Multiple
Data) and vector processing for increased efficiency.
• Task-Level Parallelism (TLP): Introduced with multi-core processors and chip
multiprocessors (CMPs), though programming and compilation challenges re-
main.
• Job-Level Parallelism (JLP): Achieved through distributed systems, allowing
large-scale applications to run concurrently across multiple nodes.
• These improvements have enabled scalable computing to handle complex com-
putations efficiently, especially in areas like high-performance computing (HPC)
and high-throughput computing (HTC).

2. Cloud and Utility Computing

• Cloud Computing: Offers scalable, on-demand computing resources over the

internet, reducing the need for dedicated infrastructure.
• Utility Computing: Adopts a business model where users pay for computing
services based on usage.
• Edge and Fog Computing: Enhance scalability by processing data closer to
the source, reducing latency and bandwidth costs.

3. Hardware and Software Innovations

• Advances in Processor Architectures: Multi-core and many-core processors

improve parallelism and computational speed.
• Scalable Storage Solutions: Distributed file systems (e.g., HDFS, Google File
System) allow efficient storage and retrieval of large datasets.
• Network Bandwidth Growth: Gilder’s Law suggests that network bandwidth
doubles yearly, supporting the increasing demand for scalable computing.

4. Emerging Technologies Driving Scalability

• Artificial Intelligence (AI) and Machine Learning (ML): Require scalable cloud
and distributed computing for training large models.
• Quantum Computing: Holds the potential for unprecedented scalability in
solving complex problems.
• Blockchain and Distributed Ledger Technologies: Enable secure and scalable
transaction processing.
• Serverless Computing: Provides scalability by dynamically managing resources
based on application demands.

Govt Engineerig College MH Halli 3 Department of CSE

Cloud Computing

Influence of Emerging Paradigms on Scalable Computing

1. Increased Resource Distribution and Parallelism
Emerging paradigms have pushed computing toward higher degrees of parallelism,
making systems more scalable and efficient. Distributed computing allows tasks to
be processed across multiple nodes, increasing fault tolerance and scalability.

2. Autonomous and Self-Organizing Systems

Scalable computing now incorporates autonomic computing, where systems self-
manage through AI-driven automation. This improves efficiency in areas like load
balancing, failure recovery, and resource allocation.

3. Software-Defined Infrastructure

• Software-Defined Networking (SDN): Enhances network scalability by allowing

centralized control over data flow.
• Software-Defined Storage (SDS): Provides scalable and flexible storage solu-
tions for cloud and big data applications.

4. Cloud-Native and Containerized Architectures

• Microservices: Enable scalable and modular application development.

• Containers (Docker, Kubernetes): Provide lightweight, portable, and scalable
application deployment.

5. Scalability in High-Performance and High-Throughput Computing

• Scalable computing is crucial for various applications across different domains,

including:
• Science and Engineering: Simulations, genomic analysis, weather forecasting.
• Business and Finance: Stock exchanges, banking, transaction processing.
• Healthcare: Telemedicine, hospital automation, disease prediction.
• Internet and Web Services: Data centers, cybersecurity, e-commerce.
• Government and Military: Crisis management, surveillance, digital gover-
nance.

6. Technology Hype Cycle and Adoption of New Innovations

The adoption of new computing paradigms follows a hype cycle, where technolo-
gies move through stages of excitement, disillusionment, and eventual widespread
adoption. Emerging trends such as cloud computing, AI, blockchain, and quantum
computing are expected to reach mainstream adoption within the next few years.

Govt Engineerig College MH Halli 4 Department of CSE

 Cloud Computing

3.Explain the role of multicore CPUs in distributed

computing and Explain about Multithreading Tech-
nology.
Answer:
Over the past 30 years, CPU processing power has increased significantly, driven by
improvements in microarchitecture, parallelism, and transistor density. Modern CPUs
incorporate multicore architectures, where a single processor chip contains multiple cores,
each capable of executing tasks independently. This evolution has enabled significant
performance gains by leveraging Instruction-Level Parallelism (ILP) and Thread-Level
Parallelism (TLP).

Figure 1: Improvement in processor and network technologies over 33 years

Multicore and Many-Core Architectures

Modern processors utilize multicore designs to improve parallel processing efficiency. A
multicore CPU consists of multiple cores, each with its private L1 cache, while L2 and
L3 caches are shared. Examples include:

1. Intel Core i7, Xeon, AMD Opteron, Sun Niagara, IBM Power 6, and X-Cell pro-
cessors.

2. The Niagara II processor has 8 cores, each supporting 8 threads, resulting in 64

parallel execution threads (8 × 8 = 64).

3. In contrast, many-core architectures are optimized for massive data-level parallelism

(DLP).

4. GPUs (Graphics Processing Units) contain hundreds to thousands of simple cores,

enabling high-performance computing (HPC) and high-throughput computing (HTC).

5. Heterogeneous computing is becoming popular, where a single chip integrates both

high-performance CPU cores and many-core GPU cores.

Govt Engineerig College MH Halli 5 Department of CSE

 Cloud Computing

Multithreading Technology
Multithreading enhances CPU utilization by allowing multiple instruction threads to be
processed simultaneously. Different multithreading models include:

1. Superscalar Processor: A single-threaded processor that executes multiple instruc-

tions per cycle using multiple functional units.

2. Fine-Grained Multithreading: Switches between multiple threads every cycle to

maximize execution unit usage.

3. Coarse-Grained Multithreading: Executes a single thread for multiple cycles before

switching to another thread.

4. Simultaneous Multithreading (SMT): Allows multiple threads to issue instructions

simultaneously to different execution units within the same cycle.

5. Chip Multiprocessing (CMP): A dual-core or multicore processor that runs multiple

threads independently across different cores.

Each architecture is optimized for different workloads, with SMT and CMP providing
the highest level of parallelism.

Figure 2: Five micro-architectures in modern CPU processors, that exploit ILP and TLP
supported by multicore and multithreading technologies.

4.Discuss the significance of GPU computing in high-performance

computing (HPC). How do GPUs improve computational effi-
ciency, and what challenges exist in their integration with tradi-
tional CPUs?
Answer:
GPUs have revolutionized high-performance computing (HPC) by providing massive par-
allelism, enabling faster computation of complex tasks. Unlike traditional CPUs, which
are optimized for sequential task execution, GPUs excel at handling thousands of parallel

Govt Engineerig College MH Halli 6 Department of CSE

Cloud Computing

threads, making them ideal for computationally intensive applications such as scientific
simulations, AI/ML, deep learning, and big data analytics.
How GPUs Improve Computational Efficiency
1. Massive Parallelism
• Modern GPUs, such as NVIDIA Fermi and Tesla GPUs, contain
hundreds or thousands of CUDA cores that process tasks concur-
rently.
• Each core can handle multiple threads, enabling the GPU to ex-
ecute thousands of operations simultaneously, significantly im-
proving performance.
2. High Throughput vs. Low Latency
• CPUs are optimized for low latency (quick execution of individual
instructions), while GPUs prioritize high throughput (processing
large datasets in parallel).
• GPUs leverage SIMD (Single Instruction, Multiple Data) and
SIMT (Single Instruction, Multiple Thread) architectures to ex-
ecute multiple computations in parallel.
3. Floating-Point Performance
• GPUs are designed to handle large-scale floating-point operations,
making them ideal for numerical simulations, weather modeling,
and deep learning.
• Modern GPUs achieve up to 82.4 teraflops (Tflops) performance,
surpassing CPUs in raw computational power.
4. Energy Efficiency
• GPUs provide higher performance per watt compared to CPUs.
• For example, a CPU core achieves less than 1 Gflop/W, while a
GPU core can reach 5 Gflops/W.
• Future Exascale computing (Eflops or 10¹ flops) systems aim for
60 Gflops/W per core, making GPUs essential for power-efficient
supercomputers.
5. Specialized Programming Models
• NVIDIA’s CUDA and OpenCL allow developers to write highly
parallelized code for GPUs.

Govt Engineerig College MH Halli 7 Department of CSE

Cloud Computing

• These frameworks enable efficient execution of AI, ML, and data-

intensive workloads.

Figure 3: The GPU performance (middle line, measured 5 Gflops/W/core in 2011),

compared with the lower CPU performance (lower line measured 0.8 Gflops/W/core in
2011) and the estimated 60 Gflops/W/core performance in 2011 for the Exascale (EF in
upper curve) in the future.

Challenges in Integrating GPUs with Traditional CPUs

Despite their advantages, GPUs face several challenges when integrated
with CPUs in HPC systems:
1. Memory Bottleneck
• GPUs rely on high-bandwidth GDDR memory, while CPUs use
low-latency DDR memory.
• Data transfer between CPU (main memory) and GPU (device
memory) creates latency overhead, limiting performance.
• Solutions include unified memory models (e.g., NVIDIA Unified
Memory) and high-speed interconnects like NVLink.
2. Programming Complexity
• Developing efficient GPU programs requires expertise in parallel
computing, thread synchronization, and memory management.
• Traditional CPU-based applications must be rewritten to leverage
GPU acceleration.
• Tools like CUDA, OpenCL, and OpenMP help, but porting legacy
code remains a challenge.

Govt Engineerig College MH Halli 8 Department of CSE

Cloud Computing

3. Limited General-Purpose Use

• GPUs excel at data-parallel tasks but struggle with branch-heavy
or sequential code.
• Certain workloads, such as database transactions and OS-level
tasks, are better suited for CPUs.
4. Power and Heat Constraints
• GPUs consume significant power, leading to thermal issues.
• Efficient cooling and energy management strategies are needed in
supercomputing clusters.
5. Concurrency and Workload Management
• Managing task scheduling between CPUs and GPUs efficiently is
complex.
• HPC systems need optimized workload distribution between pro-
cessors to avoid performance bottlenecks.

5.What is Virtualization? Explain Architectures of Virtual ma-

chines and VM Primitive Operations.
Answer:
• Virtualization is the process of creating a virtual version of a physical
computing resource, such as a server, storage device, network, or even
an operating system. It enables multiple virtual instances to run on
a single physical machine, improving resource utilization, flexibility,
and management.
• Traditionally, computers operate with a single OS tied to the hard-
ware. However, virtualization allows multiple operating systems (OS)
to run on the same hardware, each within its own Virtual Machine
(VM). A Virtual Machine Monitor (VMM) or hypervisor is used to
manage these VMs and allocate resources dynamically

Govt Engineerig College MH Halli 9 Department of CSE

Cloud Computing

Architectures of Virtual Machines

Virtual machines can be implemented using different architectures. Figure

4 illustrates three primary VM architectures:

Figure 4: Three VM architectures in (b), (c), and (d), compared with the traditional
physical machine shown in (a).

1. Native (Bare-Metal) VM
• The VMM (hypervisor) runs directly on the hardware, managing
VMs without requiring a host OS.
• Guest operating systems run on top of the hypervisor.
• Example: Xen, VMware ESXi, Microsoft Hyper-V
• Advantages: Efficient, provides direct hardware access, high per-
formance
• Disadvantages: Requires specialized hypervisor software
2. Hosted VM
• The VMM runs as an application on a traditional operating sys-
tem.
• Guest OS runs on top of the VMM, which depends on the host
OS for resource management.
• Example: VMware Workstation, VirtualBox
• Advantages: Easy to install and manage
• Disadvantages: Performance overhead due to dependency on the
host OS

Govt Engineerig College MH Halli 10 Department of CSE

Cloud Computing

3. Dual-Mode VM
• The VMM is split into user mode and privileged mode.
• Some virtualization functions are handled by the host OS, while
others run in a separate VMM layer.
• The host OS may need modifications to support virtualization.
• Example: Paravirtualization (e.g., Xen in paravirtualized mode)
• Advantages: Efficient resource utilization
• Disadvantages: Requires OS modifications

VM Primitive Operations
Virtual Machine Monitors (VMMs) enable various operations that help in
managing VMs efficiently. These operations improve flexibility, portability,
and resource utilization.
1. Multiplexing
• Multiple VMs share the same hardware resources.
• The VMM allocates CPU, memory, and I/O dynamically.
• This allows better utilization of physical resources.
2. Suspension
• A VM can be paused and saved to storage (checkpointing).
• The state of the VM, including memory, CPU registers, and disk
data, is preserved.
• This is useful for backup, migration, and fault tolerance.
3. Provisioning (Resume)
• A previously suspended VM can be resumed on the same or a
different hardware platform.
• This allows workload balancing and disaster recovery.
4. Live Migration
• A running VM is moved from one physical machine to another
without downtime.
• Useful for load balancing, maintenance, and reducing downtime
in cloud environments.

Govt Engineerig College MH Halli 11 Department of CSE

 Cloud Computing

6.What is a Computational cluster? Explain its architecture, sin-

gle system image (SSI), middleware support, and the key chal-
lenges in cluster design.
Answer
1. Introduction to Computational Clusters
• A computational cluster is a collection of interconnected stand-
alone computers (nodes) that work together to act as a unified
computing resource. These clusters leverage parallel processing
to handle heavy workloads efficiently. They are widely used in
high-performance computing (HPC), scientific simulations, and
enterprise applications.
• Clusters are built using SAN (Storage Area Network), LAN (Lo-
cal Area Network), or WAN (Wide Area Network) interconnec-
tions and can scale to thousands of nodes. Large-scale clusters
often serve as the foundation for computing grids and cloud com-
puting platforms.
2. Cluster Architecture
A typical server cluster consists of multiple compute nodes connected
via a high-bandwidth, low-latency interconnection network. The net-
work could be:
• Ethernet (standard networking)
• Myrinet (specialized low-latency network)
• InfiniBand (high-speed, high-throughput communication)
• The cluster is connected to the Internet via a VPN (Virtual Pri-
vate Network) gateway, which helps locate the cluster in a net-
worked environment.
• Characteristics of Cluster Architecture
• Loosely Coupled Nodes: Each node runs its own operating system
(OS).
• Interconnection Network: Uses Gigabit Ethernet, Myrinet, or In-
finiBand switches in a hierarchical structure.
• Resource Sharing: Nodes share I/O devices, disk arrays, and net-
work interfaces.

Govt Engineerig College MH Halli 12 Department of CSE

Cloud Computing

• Scalability: Can be expanded by adding more nodes without sig-

nificantly impacting performance.

Figure 5: A cluster of servers interconnected by a high-bandwidth SAN or LAN with

shared I/O devices and disk arrays; the cluster acts as a single computer attached to the
Internet.

3. Single-System Image (SSI)

An ideal cluster should provide a Single-System Image (SSI) to users,
meaning that multiple interconnected nodes appear as a single, uni-
fied system. This abstraction is crucial for ease of management and
performance optimization.
How SSI Works
• Resource Sharing: Merges CPU, memory, and I/O resources across
nodes.
• Transparency: Users see a single system rather than multiple
independent computers.
• Software Implementation: Achieved through middleware or spe-
cialized cluster OS extensions.
• Without SSI, a cluster is simply a collection of individual comput-
ers. SSI simplifies application deployment, load balancing, and
fault tolerance.
4. Middleware Support for Clusters
Clusters rely on specialized middleware to manage resources, sched-
ule jobs, and enable parallel processing. Middleware enables efficient
resource utilization, fault tolerance, and job scheduling.
Key Middleware Components

Govt Engineerig College MH Halli 13 Department of CSE

Cloud Computing

• Message Passing Interface (MPI): Facilitates communication be-

tween nodes in parallel applications.
• Parallel Virtual Machine (PVM): Allows heterogeneous systems
to function as a single parallel computer.
• Cluster Resource Management: Tools like LSF (Load Sharing
Facility) and Codine handle job scheduling and load balancing.
• Distributed Shared Memory (DSM): Allows multiple nodes to
share memory across the cluster.
• Middleware plays a crucial role in achieving high availability (HA),
parallel execution, and effective resource utilization.
5. Key Challenges in Cluster Design
Building and managing a computational cluster comes with several
technical and operational challenges: article array booktabs
Design Issue Challenge Feasible Solutions
Availability and Fault Ensuring uptime and handling Failover, rollback recovery, RAID
Tolerance node failures storage
Efficient Communica- Reducing message-passing la- Enhanced MPI libraries, active
tion tency message
Load Balancing Distributing workloads evenly Process migration, job replication
across nodes
Scalability Maintaining performance while Hierarchical networking, scalable
adding nodes job scheduling
Job Management Effectively scheduling and moni- Global management systems like
toring jobs LSF

Table 2: Challenges and Feasible Solutions in High-Performance Computing

One of the biggest limitations is the lack of a cluster-wide OS. Current

clusters rely on middleware and OS extensions to manage resources
effectively.

7.What is a Peer-to-Peer (P2P) system? Explain its architecture,

the concept of overlay networks, and discuss the key challenges
in P2P computing.
Answer:
A Peer-to-Peer (P2P) system is a decentralized network where each par-
ticipant (peer) acts as both a client and a server. Unlike traditional client-
server models, where a central server manages resources and services, P2P

Govt Engineerig College MH Halli 14 Department of CSE

Cloud Computing

systems rely on direct peer-to-peer interactions, making them more scal-

able, resilient, and self-organizing.
• Architecture of P2P Systems
A P2P system operates at two abstraction levels:
1. Physical Network Level:
– Peers (client machines) are connected to the Internet without
a dedicated interconnection network.
– Each peer joins and leaves the network autonomously, making
the system dynamic.
– Unlike cluster or grid computing, there is no central coordi-
nation or global control.
2. Logical Overlay Network Level:
– The logical structure of a P2P system is formed by mapping
peer IDs in a virtual network.
– Peers communicate based on their logical links rather than
their physical network connections.
– The overlay network ensures that data is distributed efficiently
and that peers can find resources.

Figure 6: The structure of a P2P system by mapping a physical IP network to an overlay

network built with virtual links.

• Overlay Networks in P2P Systems

Since P2P networks lack a central server, overlay networks are used
to establish logical connectivity among peers. There are two types of
overlay networks:

Govt Engineerig College MH Halli 15 Department of CSE

Cloud Computing

1. Unstructured Overlay Networks

– Nodes (peers) are randomly connected.
– Message passing is based on flooding (broadcasting queries to
multiple peers).
– Example: Gnutella and early P2P file-sharing networks.
– Challenges: High network traffic, inefficient search results,
and increased latency.
2. Structured Overlay Networks
– Peers are connected using a specific topology (e.g., Distributed
Hash Tables - DHTs).
– Queries follow a defined routing algorithm, making searches
deterministic and efficient.
– Example: Chord, Pastry, and CAN (Content Addressable
Network).
– Advantages: Reduced network overhead, faster data retrieval,
and improved scalability.

Key Challenges in P2P Computing

Despite its advantages, P2P computing faces several challenges:
1. Heterogeneity: Different hardware architectures, software versions,
and network protocols make interoperability difficult.
2. Scalability and Performance: P2P networks must efficiently han-
dle millions of peers and balance network traffic while maintaining
low latency.
3. Security and Trust Issues: Lack of a central authority leads to:
• Malware propagation (e.g., viruses and worms).
• Data integrity issues (no verification of shared files).
• Privacy concerns (anyone can access shared resources).
4. Data Location and Routing Efficiency: Finding a specific file
in a large unstructured P2P network requires broadcasting, which
increases network overhead.
5. Fault Tolerance and Load Balancing:

Govt Engineerig College MH Halli 16 Department of CSE

Cloud Computing

• Since peers join and leave dynamically, ensuring system stability

is challenging.
• Load balancing is difficult, as some peers contribute more re-
sources than others.
6. Legal and Ethical Concerns: P2P networks are widely used for
file sharing, leading to copyright violations and legal issues.

8.What is I/O virtualization, and how does it improve Virtual

Machine Performance?
Answer:
I/O virtualization involves managing the routing of I/O requests between
virtual devices and shared physical hardware. It ensures that multiple
virtual machines (VMs) can share physical devices efficiently. At present,

Figure 7: Device emulation for I/O virtualization implemented inside the middle layer
that maps real I/O devices into the virtual devices for the guest device driver to use.

there are three major approaches to I/O virtualization:

1. Full Device Emulation
• The Virtual Machine Monitor (VMM) replicates real-world de-
vices in software.
• The guest OS interacts with the virtual device as if it were real
hardware.
• The VMM traps all I/O access requests and processes them ac-
cordingly.
2. Para-Virtualization A more efficient I/O virtualization technique
used in Xen Hypervisor.

Govt Engineerig College MH Halli 17 Department of CSE

Cloud Computing

Uses a split driver model, where:

• The frontend driver (in the guest OS) handles I/O requests.
• The backend driver (in the privileged domain, e.g., Domain 0)
manages the real device.
3. Direct I/O Virtualization
• The VM directly accesses physical I/O devices.
• Provides near-native performance without heavy CPU usage. Self-
Virtualized I/O (SV-IO)
• Utilizes multicore processors to offload I/O virtualization tasks.
• Creates Virtual Interfaces (VIFs) for devices like virtual NICs,
virtual disks, and cameras.
• Uses message queues to manage incoming and outgoing device
communication.
• Used primarily for networking in mainframes.
• Communication happens through shared memory.

I/O Virtualization Improves VM Performance:

1. Efficient Resource Sharing: Allows multiple VMs to use a single
physical device.
2. Performance Optimization: Para-virtualization and direct I/O
improve speed compared to full device emulation.
3. Reduced CPU Overhead: Direct I/O and hardware-assisted vir-
tualization (Intel VT-d) lower CPU usage.
4. Scalability: Self-Virtualized I/O (SV-IO) enhances performance by
distributing tasks across multiple cores.

9.What are Virtual Clusters? and how do they compare to Phys-

ical Clusters.
Answer:
A virtual cluster is a group of virtual machines (VMs) that are intercon-
nected through a virtual network and run on one or more physical clusters
(groups of physical servers). These virtual clusters provide a flexible and

Govt Engineerig College MH Halli 18 Department of CSE

Cloud Computing

scalable environment for cloud computing, allowing resources to be allo-

cated dynamically based on demand.

Virtual Clusters vs. Physical Clusters

0.1 Virtual Clusters:
• Virtual clusters are created using Virtual Machines (VMs) running
on physical servers across one or more physical clusters.
• These VMs are interconnected logically through a virtual network,
spanning multiple physical networks.
• Virtual cluster boundaries are flexible and can change dynamically.
• They provide dynamic provisioning, allowing VMs to be added, re-
moved, or migrated as needed.
• VMs can run different operating systems (guest OS) than their host
machines.
• Multiple VMs can run on a single physical server, improving resource
utilization and flexibility.
• Virtual clusters support fault tolerance, disaster recovery, and dis-
tributed parallelism by replicating VMs across multiple physical servers.
• The size of a virtual cluster can scale up or down dynamically, similar
to how P2P networks change over time.
• Virtual machines can be migrated between physical nodes for load
balancing and energy efficiency (green computing).

0.2 Physical Clusters:

• A physical cluster consists of a group of physical machines (servers)
connected to work as a single system.
• These machines have fixed resources and are not as flexible as virtual
clusters.
• Each physical machine runs its own OS, and workloads are tied di-
rectly to the physical hardware.

Govt Engineerig College MH Halli 19 Department of CSE

Cloud Computing

• Expanding or shrinking a physical cluster requires adding or removing

actual hardware, which is costly and time-consuming.
• Physical clusters lack flexibility in resource allocation compared to
virtual clusters.
• Load balancing and resource management are more challenging with-
out virtualization.

Figure 8: A cloud platform with four virtual clusters over three physical clusters shaded
differently.

10.Explain the process of Live VM Migration and four ways to

manage a Virtual Cluster.
Answer:
Live Virtual Machine (VM) migration enables the movement of a running
VM from one physical host to another without significantly interrupting its
operation. This process is crucial in cloud computing and virtual clusters
for load balancing, disaster recovery, and system maintenance.
Steps in Live VM Migration:
1. Step 0: Pre-Migration
• The VM runs normally on Host A.
• A destination physical host (Host B) is selected.
• Storage devices are mirrored, and free resources are maintained.
2. Step 1: Reservation

Govt Engineerig College MH Halli 20 Department of CSE

Cloud Computing

• A container is initialized on the destination host (Host B).

• The migration process is prepared.
3. Step 2: Iterative Pre-Copy
• The VM’s memory is copied to Host B while the VM is still
running.
• The system tracks ”dirty pages” (memory pages that change dur-
ing migration).
• These dirty pages are copied iteratively until they are small enough
to be transferred in one final step.
4. Step 3: Stop-and-Copy
• The VM is suspended on Host A (causing a short downtime).
• The final state of memory, CPU, and network data is transferred
to Host B.
• An ARP (Address Resolution Protocol) request is generated to
redirect network traffic.
5. Step 4: Commitment
• Host A releases the VM state.
• The migration process finalizes.
6. Step 5: Activation
• The VM starts running on Host B.
• The network and storage connections are re-established.

Govt Engineerig College MH Halli 21 Department of CSE

Cloud Computing

Figure 9: Live migration process of a VM from one host to another.

Four Ways to Manage a Virtual Cluster:

1. Guest-Based Cluster Management
• The cluster manager resides on the guest systems (VMs).
• Multiple VMs form a virtual cluster.
• Example: openMosix (Linux cluster on Xen), Sun’s Cluster Oasis
(Solaris cluster on VMware).
• Pros: VMs can be easily moved and replaced.
• Cons: Host system failures can impact VMs unless live migration
is enabled
2. Host-Based Cluster Management
• The cluster manager runs on the host system, which supervises
the VMs.
• The host system can restart VMs on another physical machine if
necessary.
• Example: VMware HA (High Availability), which restarts VMs
after failure.
• Pros: Ensures reliability and failover at the host level.
• Cons: Less flexible for VM-level customizations.

Govt Engineerig College MH Halli 22 Department of CSE

Cloud Computing

3. Independent Cluster Management (Hybrid Approach)

• Cluster management runs both on host and guest systems.
• Provides better control but increases complexity.
• Pros: Balances VM flexibility and host-level control.
• Cons: More complex infrastructure and management overhead.
4. Integrated Cluster Management
• The cluster manager is designed to distinguish between physical
and virtual resources.
• Allows seamless management of both physical and virtualized in-
frastructures.
• Pros: Most efficient in large-scale cloud computing and HPC sys-
tems.
• Cons: Requires advanced orchestration tools and more computing
power.

11.What is Server Consolidation, and how does virtualization

help reduce hardware costs in data centers?
Answer:
Server consolidation is the process of reducing the number of physical
servers in a data center by combining workloads onto fewer, more powerful
machines using virtualization. Instead of running separate applications on
multiple underutilized servers, organizations use virtual machines (VMs)
to host multiple workloads on a single physical server.
Virtualization reduces hardware costs in data centers by maximizing
resource utilization and minimizing the need for physical servers.
Key Ways Virtualization Lowers Costs:
1. Increased Server Utilization: Traditional servers operate at 10-15
percentage of utilization, while virtualization allows 70-80 percentage
of utilization, reducing hardware requirements.
2. Reduced Physical Hardware: By running multiple VMs on a sin-
gle server, organizations buy fewer physical servers.
3. Lower Power and Cooling Costs: Fewer servers mean less energy
consumption and lower cooling costs.

Govt Engineerig College MH Halli 23 Department of CSE

Cloud Computing

4. Decreased Maintenance Costs: With fewer physical machines,

hardware maintenance and upgrade costs are reduced.
5. Efficient Disaster Recovery: Virtualization simplifies backup and
recovery, reducing costs associated with downtime and redundant
hardware.

12.How virtualization is implemented in Data Center Automa-

tion? What is meant by Server Consolidation?
Answer:
Virtualization in data center automation is implemented by dynamically
allocating hardware, software, and database resources to users while en-
suring Quality of Service (QoS) and cost-effectiveness. The automation
process is enabled through virtualization products and cloud computing
services. Key aspects of virtualization in data centers include:
1. Server Virtualization: Physical servers are divided into multiple
virtual machines (VMs), allowing efficient use of resources.
2. Virtual Storage Management: Storage is virtualized using tech-
niques like Parallax to provide scalable and isolated storage for VMs.
3. Dynamic Resource Scheduling: VMs are allocated resources dy-
namically based on demand, ensuring optimal performance and power
efficiency.
4. High Availability and Backup Services:Virtualization enables
quick migration of VMs between servers, enhancing reliability and
disaster recovery.
5. Cloud Operating Systems (Cloud OS): Virtualization managers
like Eucalyptus, OpenNebula, and vSphere help manage virtual in-
frastructure, automating VM provisioning and resource allocation.
Server consolidation is the process of reducing the number of physical
servers in a data center by running multiple virtual machines (VMs) on
fewer servers. This improves resource utilization and reduces hardware,
power, and cooling costs.

Govt Engineerig College MH Halli 24 Department of CSE

Cloud Computing

13.How Trust Management in Virtualized Data Centers is done?

Answer:
A Virtual Machine Monitor (VMM) introduces a layer of software between
the operating system and hardware, allowing multiple Virtual Machines
(VMs) to run on a single physical system. This encapsulation of VM states
presents both security benefits and challenges, making trust management
a critical aspect of virtualized data centers.
Key Aspects of Trust Management:
1. VMM as the Security Base
• The VMM controls hardware resource access, ensuring isolation
between VMs.
• A management VM (privileged VM) can create, suspend, resume,
or delete VMs.
• If an attacker gains control over the VMM or management VM,
the entire system is at risk.
2. Challenges in VM Security
• VM state encapsulation allows cloning and sharing, increasing
the risk of unauthorized access.
• Rollback attacks: If a VM is rolled back after generating a random
session key, it may reuse the same key, leading to encryption
vulnerabilities.
• TCP hijacking: Reuse of TCP sequence numbers can lead to
session hijacking.
VM-Based Intrusion Detection:
Intrusion Detection Systems (IDS) are essential for recognizing unau-
thorized access. IDS can be classified into:
• Host-Based IDS (HIDS): Runs on the monitored system but may
be compromised during an attack.
• Network-Based IDS (NIDS): Analyzes network traffic but cannot
detect certain local attacks.
Virtualization-Based IDS
• Provides better isolation between VMs, preventing attacks from spread-
ing.

Govt Engineerig College MH Halli 25 Department of CSE

Cloud Computing

• VMM-based monitoring tracks access requests, reducing the chances

of false actions.
Implementation Methods:
• Dedicated IDS VM: Runs as an independent high-privileged VM
on the VMM.
• Integrated IDS in VMM: Has full access to hardware, enhancing
detection.
Example: Livewire IDS Architecture
• Uses a policy engine and policy module to monitor VM activities.
• Implements PTrace for tracking security policies.

Figure 10: The architecture of livewire for intrusion detection using a dedicated VM.

Trusted Zones for Virtual Clusters (EMC VMware Example)

EMC and VMware developed security middleware to manage trust in
virtualized environments. Trusted Zones ensure VM security by:
1. Segregating tenant environments using isolated virtual clus-
ters.
2. Applying security mechanisms, including:
• Anti-malware intrusion detection.
• Access control identity federation.
• Data encryption loss prevention.
• End-to-end security compliance

Govt Engineerig College MH Halli 26 Department of CSE

Cloud Computing

Figure 11: Techniques for establishing trusted zones for virtual cluster insulation and
VM isolation.

Govt Engineerig College MH Halli 27 Department of CSE