Cloud Computing

Shajulin Benedict
[email protected]
Indian Institute of Information Technology Kottayam
www.sbenedictglobal.com
1.1 Cloud Computing - Introduction

www.sbenedictglobal.com
Cloud Computing - Industries

Company
wants
the data in
Cloud…

www.sbenedictglobal.com
What is Cloud Computing

Different views and perspectives…

Think of the elephant story….

https://en.wikipedia.org/wiki/Blind_men_and_an_elephant

www.sbenedictglobal.com
What is Cloud Computing? - Definition
• According to Forrester,
• Cloud computing is a form of standardized IT-based capability
– such as Internet-based services, software, or IT
infrastructure – offered by a service provider that is accessible
via. Internet protocols from any computer is always available
and scales automatically to adjust to demand, is either pay-
per-use or advertising-based has Web- or programmatic
based control interfaces, and enables full customer self-
service.
• According to NIST
• Cloud Computing is an approach to computing pool of
configurable computing resources (e.g., networks, servers,
storage, applications, and services) that can be rapidly
provisioned and released with minimal management effort or
service provider interaction.
5

www.sbenedictglobal.com
What is Cloud Computing? - Definition
• According to Buyya
• Cloud is a parallel and distributed computing system
consisting of a collection of inter-connected and virtualized
computers that are dynamically provisioned and presented as
one established through negotiation between the service
providers and consumers.
• According to McKinsey Co
• Clouds are hardware based services offering compute,
network, and storage capacity where: Hardware management
is highly abstracted from the buyer; buyers incur infrastructure
costs as variable ones; and, infrastructure capacity is highly
elastic.

www.sbenedictglobal.com
 Cloud Computing and Services

Own IT IaaS Cloud PaaS Cloud SaaS Cloud

Amazon WS
Salesforce.com
SeverPath Windows Azure
Google Apps
appNexus Google App Engine
Zoho Office
Eucalyptus

Data Data Data Data

Application Logic Application Logic Application Logic
Appl. Logic
Devel./Runtime Devel./Runtime Devel./Runtime Devel./Runtime
Infrastructure Infrastructure Infrastructure Infrastructure

Will be repeated in future sessions…

www.sbenedictglobal.com
Who needs Cloud?
• Almost Everyone!!!

Pay as you go
Developers
Or
Users

www.sbenedictglobal.com
History – Cloud Computing

“A computing paradigm where the boundaries of computing will be

determined by economic rationale rather than technical limits alone.” –
Prof. Ramnath Chellapa..

www.sbenedictglobal.com
Challenges
• Scalability issues
• Inefficient most of the times – replication
• Data security
• Ownership, proprietary, trust, …
• Fault tolerance issues
• Green issues
• Data management issues – locality, zones, replication

www.sbenedictglobal.com
1.2 Base Technologies – Cloud Computing
• Contents (5 Parts):
Computing Domains

CPU,GPU,Memory

1.2 Base
Technologies

Technology Scalable system

Convergence Models

Virtualization

www.sbenedictglobal.com
1.2.1 Computing Domains

Computing Domains
–Evolutionary changes – HTC / HPC
– Age of Internet computing

www.sbenedictglobal.com
 Evolutionary Changes in Computing

Age of Internet Computing

Changes – Machine architecture, OS, network connectivity, and application scenario

www.sbenedictglobal.com
Computing distinctions
Central Parallel Distributed Cloud
computing Computing Computing Computing
Computer All processors Here, The resources
resources are are tightly computers that can be either a
centralized in coupled with have private centralized or
one physical centralized memories are distributed
system. shared memory. connected and computing
communicated system.
through a
network. A form of utility
computing or
service
computing.

www.sbenedictglobal.com
Age of Internet Computing
• Billions of people use the internet.
• Thus, internet users require HPC capabilities at their
desk.
• The emergence of clouds indeed requires HTC built
over parallel and distributed computing technologies.

www.sbenedictglobal.com
Generations
• Computer technology has reached five generations!
• Each generation had 10 to 20 generations
• Overlap of around 10 years.
• 1950 – 70 → mainframes
• 1960 – 1980 → lower cost mini computers
• 1970 – 1990 → PCs with VLSI microprocessors
• 1980 – 2000 → wired and wireless massive portable
computers
• Since 1990 → HPC and HTC systems emerged (in the form of
clusters, grids, clouds ..)

www.sbenedictglobal.com
Evolution of HTC and HPC over Internet

Virtualization
Characteristics
Of HTC and HPC RFID and sensors

HPC -> Supercomputers are replaced by clusters of computers

HTC -> P2P networks, clouds, and webservice platforms are more
focused on HTCs.
17

www.sbenedictglobal.com
Scalable Computing trends
• Several factors drive computing applications
• Moores law
• number of transistors per square inch on integrated circuits had
doubled every two year (WIKI),
• processor speed doubles every 18 months.
• Gilders law
• bandwidth grows at least three times faster than computer power”
• Price/performance ratio of systems
• refers to a product's ability to deliver performance
• Degree of parallelism
• indicates how many operations can be or are being
simultaneously executed by a computer.

www.sbenedictglobal.com
Degree of Parallelism
• Degree of parallelism
• bit-serial processing fashion (50 years ago)
• bit-level parallelism – word serial processing
• instruction-level parallelism (ILP) – due to 8-,16-,32-64bit
procs.
– Used in modern CPUs – demand h/w and compiler support
– Multiple issue superscalar architecture,
– Dynamic branch prediction
• data-level parallelism – explored in GPUs (many cores)
• task-level parallelism – explored in GPUs (many cores)
• job-level parallelism

www.sbenedictglobal.com
Scientific Applications – Possibility to Scale
• Science and Engineering
• Simulations, genomic analysis, seismic, earthquake
simulations, weather forecasting, HEP
• Business, Education, and Health care services
• Content delivery, transaction processing, distance education..
• Internet and web services
• Traffic monitoring, digital government, online tax return
processing.
• Mission critical applications
• Crisis managements, recovery, intelligent systems. ..

www.sbenedictglobal.com
Application Requirements

• These applications demand

• Economics
• Scalability
• QoS
• System reliability
• Security and so forth.
• Transparency to end users and system managers
– In terms of data access, resource allocation, process location,
concurrency, job replication, and failure recovery

www.sbenedictglobal.com
1.2 Base Technologies – Cloud Computing

Computing
Domains

CPU,GPU,Memory

1.2 Base
Technologies

Technology Scalable system

Convergence Models

Virtualization

www.sbenedictglobal.com
1.2.1 CPU-GPU-Memory-Networks

• Development of HPC or HTC systems require

sophisticated growth in component and network
technologies.
• Multicore vs. Multithreading
• GPU computing
• Memory, storage, WAN

www.sbenedictglobal.com
Multicore CPU Technology
• Recent processors are having n processing cores
(dual, quad, six, eight, and more)
• The manufacturers try to reduce the size of the
processors and increase the core count.
• Intel has introduced tick-tock model since 2007.
• Tick / Tock duration → 12 to 18 months
• Tick duration → shrink the size of processors
• New fabrication process
• Tock duration → frame new architecture
• New micro-architecture

www.sbenedictglobal.com
Multicore technology – contd.
Tick
Model Processor Name Size Year

Tock
/ Nahalem 45 nm 2008

Tock Sandy Bridge 32 nm 2010

Tock
Tick Ivy Bridge 22 nm 2011
Model
Tock Haswell 22 nm 2013

Tick Broadwell 14 nm 2014

Tock Skylake 14 nm 2015

Tick** Kaby Lake 14 nm Q3-2016

Q3-2017 (Delayed—
Cannon Lake 10 nm
2019?

Icy Lake 10 nm 2018 -- 2019 ?

Tiger Lake 10nm 2020

Alder Lake 10nm 2021

Lunar Lake Didn’t talk about size 2025

www.sbenedictglobal.com
Multicore Technology – Generic View.

• A typical multicore architecture consists of L1, L2, L3

Multicore CPUs may increase
from tens of cores to hundreds of
Core 1 Core 2 Core n
cores in future.
L1 L1 L1
Many-core GPUs has > 100s of
thin cores.
L2 cache
Caches and memories

L3 cache / DRAM

26

www.sbenedictglobal.com
Multi-core technology – contd.
• The clock rate got increased
• 10 MHz for Intel 286 to 6 GHz for 14th generation Intel Rapter
Lake (20 cores too)
• It is constantly being increased until last decade
• Clock rate growth limitation
• Immature chip manufacturing technology reduced the growth.
• Too much of heat dissipation with high frequencies or high
voltages.
• Thus, many core CPU evolved.

www.sbenedictglobal.com
Manycore GPU
• GPUs are involved.
• GPU is a graphics coprocessor (accelerator) mounted on a
computers graphics card or video card.
• CPU instructions are offloaded from CPU to GPUs.
• First GPU was NVIDIA’s GeForce 256 (1999)
• HPC community uses GPGPUs (General Purpose computing
on GPUs).
• Top 500 lists have many GPGPU systems.
• Typical GPGPU:
• NVIDIA GPUs may have 128 GPU cores on a chip.
• Each core may handle 8 threads of instructions.
• Thus, nearing 1024 threads of parallelism.
• Data intensive calculations are offloaded from CPU and
executed in parallel on these threads.
• Multiple GPUs could be clustered to form a massive parallel 28
GPU. www.sbenedictglobal.com
Skylake Architecture

www.sbenedictglobal.com
Intel Kabylake Processor

www.sbenedictglobal.com
Xeon Phi – Knights Landing
Intel Xeon Phi generations
Knights Corner, 2012
Knights Landing, 2016
Knights Hill

www.sbenedictglobal.com
Quantum Computers
Intel, IBM, Google, Microsoft and the other giants are
developing quantum computers.

Billions of dollars are invested on quantum computers.

Quantum Computing is better than classical computers.

• It applies quantum mechanics.
• Speedup outperforms.
• Deals with quantum dynamics of electrons.
• Information is represented using qubits.
• Microscopic particles such as photons, electrons, atoms,
and ions are represented in qubits. (These particles are
controlled by devices such as ion traps and quantum dots).

www.sbenedictglobal.com
Quantum Computing -- History
Quantum computing
1981 – Conference took place by MIT and IBM
1995 – Error correction using quantum physics
2016 – IBM Q experience launched.
Key question of the conference?
Relationship between physics and information…
Key theme:
Simulate information based on nature…

www.sbenedictglobal.com
Quantum Computing

www.sbenedictglobal.com
Quantum Computing - Qubits
Classical computers deal with 1 or 0 (two states)
QC deals with 1 or 0 or both at the same time.
QC can stay in the superposition state (i.e.,
simultaneously in 1 and 0)

Qubits are different from bits!

Entanglements
Superposition
Interference

www.sbenedictglobal.com
Qubits - Kinds
Spin of electrons
Based on magnetic fields (spin up / down)
Atoms energy level
High, Low energy levels
Photons
Horizontal or vertical polarization based on light
Superconducting circuits
At very low temperature, some materials flow current without
any resistance.
(Clockwise / Anticlockwise – and, atoms).

https://www.youtube.com/watch?v=KhrTTqwKjn4

www.sbenedictglobal.com
Quantum Computer
How does it look like?

IBM-Q:

www.sbenedictglobal.com
Memory, Storage, and WAN
Year DRAM Year Disk
Memory Capacity
Chip (Hard Disk)
capacity
1976 16KB 1981 260MB
2011 64GB 2011 3TB (Seagate
barracuda XT)

• Memory is an essential component for obtaining good gaming experience!!

• 4-8 GB is sufficient for normal works.
• But, for gaming, typically we need 16 GB

• Memory chip capacity got a 4x increase in every three years.

• Disk array capacity got a 10x increase in every 8 years.

• The memory access time did not improve….

• But, memory capacity and processor speed got increased…

https://www.youtube.com/watch?v=DLM20pWqMyU

www.sbenedictglobal.com
Memory, Storage, and WAN
• Storage Technology
• Rapid growth in flash memory and Solid State Drives (SSDs)
• NAND flash (non volatile)
• Emerging memory technologies
• Nanowire based storage
• These growth factors impact the future of HPC or HTC.
• Typically, SSD can handle 300000 to 1 million write cycles per
block.
• HDD – Hard Disk Drive – (in 2022, 30TB HDD!)
• Tape units are dead!!!
• Disks are tape units!!!
• Flashes are disks!!!
• And, memory are caches!!! – now…

• 39

www.sbenedictglobal.com
Memory, Storage, WAN
• System Interconnects

More
Concurrent
Computers
are required

cConnected via.
Internet in a
Hierarchical fashion

www.sbenedictglobal.com
1.2 Base Technologies – Cloud Computing

Computing
Domains

Networking
Systems

1.2 Base
Technologies

Scalable
Technology
system
Convergence
Models

Virtualization

www.sbenedictglobal.com
1.2.3 System Models

System Models for Scalable Cloud

Computing

www.sbenedictglobal.com
Scalable Computing – Basics

Cluster computing
Computational grids
P2P networks Cloud
Internet technologies Computing
Virtualization

Underlying Scalable
Technologies Cloud
www.sbenedictglobal.com
Scalable Cloud Computing - Models
• A gigantic computer with an elephantine specification
might not solve complex problems.
• We need scalable distributed computing solutions.
• These scalable systems involve 100s, 1000s, or even
millions of computers.
• These computers are
• Collectively
• cooperatively,
• or collaboratively involved
at various levels of
executions.

www.sbenedictglobal.com
Cluster Computing - Architecture

Disks
And
Server I/O
Ethernet or Myrinet or Internet via.
Infiniband or high speed n/w Gateway

• Computer nodes are interconnected using a low-

latency, high bandwidth interconnection network.
• Gigabit Ethernet (10Gbps)
• Myrinet from Myrion
• and Infiniband networks
• Intel and Mellanox
• 54 to 96 Gbps with 4 links
Cluster Computing
• Using hierarchical constructions, we could create a
scalable cluster.
• They are also called a loosely coupled parallel
computers which share a file system.
• Usually, they have multiple system images
• This results in many autonomous nodes under different OS
control.
• Single System Image would be an ideal Cluster.
• SSI presents a collection of resources as one integrated,
powerful resource.

www.sbenedictglobal.com
Cluster Computing
• Benefits
• Scalable performance
• Efficient Message passing
• High system availability
• Seamless fault tolerance
• Cluster-wide job management
• Challenges
• A dedicated OS to operate cluster as SSI is not available.
• S/w and applications must rely on the middleware to achieve
high performance.

www.sbenedictglobal.com
Supercomputer
• System:
• These are a large collection of clusters housed in a single
building. Cloud do utilize those supercomputers to frame
datacenters.
• The computing units are tied up with high speed networks.
• OS:
• Earlier, supercomputers had single OS (such as, Chippewa
OS – mostly, as a job sharing system).
• The software costs increased tremendously. And, the
designers switched to unix style OS.
• Modern Supercomputer OS
• Each nodes were installed with different versions of OS.
• Drawbacks of Supercomputers
• Expensive ones.
www.sbenedictglobal.com
SuperMUC @ Leibniz Supercomputer Centre

Courtesy of Prof. Gerndt, Yury Oleynik, and LRZ scientists

www.sbenedictglobal.com
LRZ Supercomputer
• Massively parallel system with more than 241,000
cores.
• 2 Phases of implementation (2011 – 2014)
• It reaches 6.8 petaflop/s with LINPACK
• RANKED – 27th in the top500 list.
• SunwayTaihulight from NRCPC – (93 PetaFLOPS)
• FLOPS – floating point operations per second (FLOPS for
supercomputers –MIPS for PCs)
Mega 106 million
Giga 109 billion
Tera 1012 trillion
Peta 1015 quadrillion
Exa 1018 quintillion
Zetta 1021 sextillion
www.sbenedictglobal.com
LRZ Supercomputer
Phase II
Phase I

Category 2011 2012 2013 2014 (Phase II)

Islandtype Fat Nodes Thin Nodes Many Cores Nodes Haswell Nodes

Processor Westmere Sandybridge Ivybridge Haswell

Total Number of
205 9216 32 3072
nodes
Total Number of
8,200 147,456 3,840 (Phi) 86,016
cores
Total Number of
1 18 1 6
Islands

Nodes per Island 205 512 32 512

Processors per 2 (IvyB) 2.6 GHz +

4 2 2
Node 2 Phi 5110P
Cores per
10 8 8 (IvyB) + 60 (Phi) 14
Processor

www.sbenedictglobal.com
Phase – I ---- LRZ SuperMUC
• A distributed memory architecture with 19 partitions
• 18 partitions called islands + 1 I/O
• Node is a shared memory system
with 2 processors
• Sandy Bridge-EP
Intel Xeon E5-2680 8C
• Sandy -> 2.7 GHz (3.5 GHz)
• 32 GByte memory
• Inifiniband network interface
• Processor has 8 cores
• 2-way hyperthreading
• 21.6 GFlops @ 2.7 GHz per core
• 172.8 GFlops per processor

www.sbenedictglobal.com
 Sandy Bridge Processor

Latency: Core 8 multithreaded cores Core

8 multithreaded cores
• Latency:
4 cycles Bandwidth:
• • 124 cycles • 2*16/cycle
cycles L1 32KB L1 32KB
• 31 cycles • 32 / cycle
• 12 cycles L2 256KB L2 256KB • 32 / cycle
• 31 cycles

L3 2.5 MB Shared L3 L3 2.5 MB Network frequency

Network frequency
equal to core frequency
equal to core frequency

Memory QPI PCIe

L3 cache
• Shared cache among two processors in a node.

www.sbenedictglobal.com
NUMA Node
4GB 2 QPI links 4GB
4GB 4GB
Sandy Bridge Each 2 GT/s Sandy Bridge
4GB 4GB
4GB 4GB

8xPCIe3.0 (8GB/s)
Infiniband

• 2 processors with 32 GB of memory

• Aggregate memory bandwidth per node 102.4 GB/s
• Latency
• local ~50ns (~135 cycles @2.7 GHz)
• remote ~90ns (~240 cycles)
www.sbenedictglobal.com
Phase – I LRZ SuperMUC

www.sbenedictglobal.com
Performance of Phase-I

Rest for fat

126 spine
36 port switch 36 port switch node and
switches
IO
19 links

126 links

648 port switch 18 islands 648 port switch

+ IO island

Nodes

www.sbenedictglobal.com
Interconnection Network
Infiniband FDR-10
FDR means Fourteen Data Rate
FDR means 14Gb/s per lane
FDR-10 has an effective data rate of 38.79 Gbit/s
Latency: 100 nsec per switch, 1usec MPI
Vendor: Mellanox
Intra-Island Topology: non-blocking tree
256 communication pairs can talk in parallel.
Inter-Island Topology: Pruned Tree 4:1
128 links per island to next level

www.sbenedictglobal.com
 MPI Performance – IBM MPI over Infiniband
Bandwidth MB/s
Bandwidth MB/s

Message length (KB)

www.sbenedictglobal.com Message length (KB)
 MPI Performance – IBM MPI over Ethernet
Bandwidth MB/s
Bandwidth MB/s

Message length (KB)

Message length (KB)
 Compute Nodes

Cold Corridoor
Infiniband (red)
and
Ethernet (green)
cabling

www.sbenedictglobal.com
Infiniband Interconnect
19 islands with 126 Spine Switches

11900 Infiniband Cables

www.sbenedictglobal.com
 IO System

Spine Infiniband Switches

GPFS for
$WORK and
$SCRATCH Login nodes $HOME Archive

10+5 PB @ 200 +100 GB/s 5 PB @ 80 Gb/s 30 PB @ 10GbE

www.sbenedictglobal.com
 SuperMUC Power Consumption
Linpack HPL run May 17,2012 – 2.582 PF
Run Start: 17.05.2012 20:56, 965,40 kW

SuperMUC HPL Power Consumption (Infrastructure, Run End: 18.05.2012 08:37, 711,02 kW
Duration: 42045s or 11.68 hours
Machine Room & PDU Measurements) Avg. power: 2758.87 kW
4000 50000 Energy: 32308.68 kWh

45000
3500
40000 Power (Machine Room,
3000 kW)

Energy Consumption (kWh)

35000 Power (PDU, kW)
2500
30000
Power (kW)

Power (infrastructure,
2000 25000 kW)
Energy (Machine Room,
20000 kWh)
1500
15000
1000
10000
500 Subsystems included in PDU
5000
measurements:
0 0
o Computational Nodes
16:301 8:001 9:302 1:002 2:300:00 1:30 3:00 4:30 6:00 7:30 9:0010:301 2:001 3:30
Time (Clock) o Interconnect Network

Provided by Torsten Wilde, LRZ

www.sbenedictglobal.com
 SuperMUC - Costs
• 1 million euro = 9 crore Rs.

2010-2014 2014-2016
Phase 1 Phase 2
High End System
Investment Costs (Hardware
53 Mio € ~ 19 Mio €
and Software)
Operating Costs (Electricity
costs and maintenance for
32 Mio € ~ 29 Mio €
hardware und software, some
additional personnel)
SUM 85 Mio € ~48 Mio €
Extension Buildings (construction
and Infrastructure) 49 Mio €

www.sbenedictglobal.com
Energy Capping in Contract with IBM
• New funding scheme: energy included
• Contract includes energy cost for 5 years
• Power consumption of the system varies between
1 and 3 MW depending on the usage by the
applications.
• Power bands
• LRZ has upper and lower power bands to utilize in the
applications.
• Contract penalty of 100000 Euros.
• The contract is based on the energy consumed in a
benchmark suite agreed between IBM and LRZ

www.sbenedictglobal.com
IBM iDataplex dx360 M4: Water cooling
• Sandy&Ivybridge processor-based machines utilized
IBM iDataPlex system (Thin nodes and manycore
nodes).
• Heat was cooled by water – Free cooling.
• Power advantage over air-cooled node:
• Warm water cooled ~10% Input=18 to 45 degree centigrade
Output = < 50 degree centigrade
(cold water cooled ~15%)
• due to lower Tcomponents and no fans
• Typical operating conditions:
• Tair = 25 – 35°C
• Twater = 18 – 45°C
• No fans
• Less energy and noise

www.sbenedictglobal.com
LRZ Infrastructure layout

https://www.youtube.com/watch?v=LzTedSh51Tw
Cooling process

Hö
Höchstleistungsrechner
(sä
(säulenfrei)
Zugangsbr
Entranceü cke

Server/Netz

Archiv/Backup
Archiv/Backup

Cooling
Klima

Electrical Systems

www.sbenedictglobal.com
P2P Networks
• P2P computing has hierarchical levels of networking –
logical and physical.
• In P2P architecture, P2P systems are introduced at
the physical level and overlay networks are at the
logical level.
• In P2P, all nodes act as a client and a server.
• All clients autonomously choose their peers (join or
leave on their own).
• No central coordination or central database is needed.

69

www.sbenedictglobal.com
P2P Architecture

Overlay
networks –
logical

Physical
network and
Computers

• Clients contribute resources for the tasks. They are involved in the
ongoing P2P computation.
• Task level parallelism could be achieved using P2P.
• Dynamic formation of topologies are possible using P2P.
70

www.sbenedictglobal.com
P2P Architecture
• Data items or files are distributed in the participating
peers.
• Based on the communication, the IDs are initialized
and overlay networks (logical) are formed.
• 2 types of overlay networks
• Unstructured – a random graph, no fixed route, flooding
• Structured – fixed connection topology, routing algorithms..
• 3 design objectives
• Data locality
• Network proximity (nearness in the space)
• Interoperability

www.sbenedictglobal.com
Scalabe P2P Grid - Vishwa
Vishwa P2P Grid – IIT Madras (Prof. Janakiram)

www.sbenedictglobal.com
P2P Application Families
• Four Application groups are identified in P2P Systems
• For distributed file sharing
– Eg. BitTorrent
– Eg. Content distribution of MP3, music, video.
• For collaborative networking
– Eg. MSN, Skype, instant messaging
• For distributed computing
– Eg. Vishwa, SETI@home [search for extra-terrestrial intelligence]
• For other purpose, such as, discovery, communication,
security, and resource aggregation
– Eg. JXTA (programming), .NET, FightingAID@home (finding new
drug for HIV)

www.sbenedictglobal.com
P2P Computing Challenges
• Hardware/Software/Network Issues
• P2P needs to handle too many hardware models and
architectures.
• Incompatibility exists between software and the OS.
• It needs to handle different network connections and
protocols. (Scalability issues).
• P2P performance is affected by routing efficiency and
self-organization by participating peers.
• As P2P is not centralized, manageability is also
difficult.

www.sbenedictglobal.com
Computing Grids
• Power grid concept.
• Resource sharing & coordinated problem solving in
dynamic, multi-institutional virtual organizations.
• Difference from clustering is that the grids are widely
spread across the globe.
• Distributed vs. Grid computing
Grid is an evolution of distributed computing
Dynamic
Geographically independent
Built around standards
Internet backbone

www.sbenedictglobal.com
Why Grids?
Large-scale science and engineering are done through
the interaction of people, heterogeneous computing
resources, information systems, and instruments, all
of which are geographically and organizationally
dispersed.

The overall motivation for “Grids” is to facilitate the

routine interactions of these resources in order to
support large-scale science and Engineering.

www.sbenedictglobal.com
Grid computing
• Special instruments may also be involved.
• Eg. Radio telescope in SETI@home
• Grid is often constructed across LAN, WAN, or
internet backbone networks at a regional, national, or
global scale.
• 3 kinds
• Computational grids
• Data grids
• P2P grids.

www.sbenedictglobal.com
Grid Architecture – Hourglass Model

User Applications

Services-Directory brokering,
monitoring

Secure access to grid resources

Grid resources – storage, fabric,

computers, clusters…..

www.sbenedictglobal.com
1.2 Base Technologies – Cloud Computing

Computing
Domains

Networking
Systems

1.2 Base
Technologies

Scalable
Technology
system
Convergence
Models

Virtualization

www.sbenedictglobal.com
Technology convergence

Computing Paradigms

Attributes / Capabilities

www.sbenedictglobal.com