Part 1 – The Era of Cloud Computing

Chapter 1 – The Motivations For Cloud

Cloud Computing Everywhere

The following are examples of situations that involve cloud computing:

• A startup leases cloud facilities for its web site; the company can pay for additional facilities as web
traffic grows.
• An individual uses a smart phone to check Internet of Things (IoT) devices in their residence.
• An enterprise company leases facilities and software for business functions, such as payroll,
accounting, and billing.
• Students working on a team project use a browser to edit a shared document.
• A patient wears a medical device that periodically uploads readings for analysis; their doctor is alerted
if a medical problem is detected.
• A seasonal company leases computing facilities during four peak months each year; the
company doesn’t pay for facilities at other times.
• A teenager logs into a social media site and uploads photos.
• A retail company leases facilities at the end of each fiscal year to run data analytics software that
analyzes sales for the year.
• An individual uses a streaming service to watch a movie; a copy of the movie is kept in a facility near
the family’s residence.
• The recipient of a package uses a tracking number to learn about the current location of the package
and the expected delivery time.

A Facility For Flexible Computing

• Many of the examples illustrate a key aspect of cloud computing: flexibility A cloud offers flexible computing
facilities (servers and software), storage facilities, and communication facilities (Internet connections) to
accommodate both incremental growth and cyclic demand.

• Cloud computing allows each customer to increase or decrease their use of cloud facilities at any time; a
customer only pays for the facilities they use.

The Start Of Cloud: The Power Wall And Multiple Core

• Two intertwined factors contributed to the start of the cloud paradigm:

o Technological: limits on speed forced a move to parallelism

o Economic: changes in technology changed IT costs

• Moore’s Law — prediction that the number of transistors would double every eighteen months.

o Result - As the size of a given transistor was shrinking, more transistors were squeezed together on a
chip each year.

o Problem: heat
 • Power wall – Chip manufacturers eventually reached a critical point, and processor speeds could not be
increased beyond a few Gigahertz without generating so much heat that the chip would burn out.

• Multicore processors - chip manufacturers devised chips that contain multiple copies of a processor (cores).
Each core consists of a complete processor that operates at a safe speed, and software must devise a way to use
multiple cores to perform computation.

From Multiple Cores To Multiple Machines

• As the number of cores increases, I/O and memory accesses become a bottleneck.

• Data grew beyond the capabilities of even the most powerful supercomputers.

• Supercomputers became too expensive but as personal computers became more powerful and less expensive; a
new form of supercomputing evolved by interconnecting a large set of inexpensive personal computers.

• The resulting configuration, known as a cluster architecture, has a key advantage: processing power can be
increased incrementally by adding additional inexpensive commodity computers.

From Clusters To Web Sites And Load Balancing

• Web sites and scientific computing systems differ in a fundamental way. Super- computer clusters are
designed so that small computers can work together on one computation at a time. In contrast, a web site must
be designed to process many independent requests simultaneously.

• Problem: how can a web site scale to accommodate thousands of users?

• Solution: A special-purpose hardware device (load balancer) that divides incoming traffic among a set of
computers that run servers

incoming web traffic server 1

load
balancer server 2

Internet company
server 3 database
.
1/N of the traffic .
goes to each server server N

Figure 1.1 Illustration of a load balancer used to divide incoming network traffic among a set of
computers.

• Load balancing technology ensures that all communication from a given customer goes to the same server.

• If the site has N servers, load balancing means that each server handles approximately 1/N of the customers.

• At any time, a site could be expanded merely by adding additional servers; the load balancer
automatically includes the new servers when dividing traffic.
Racks Of Server Computers

• Computers reduced in size so that multiple server computers can be stacked vertically into tall metal
equipment cabinets called racks.

• A full-size rack is approximately six and one-half feet tall, two feet wide, and three and one-half feet deep.

• The rack contains metal mounting bars called rails to which equipment can be attached.

• A rack is not full of servers. At least one slot in a rack (usually the top of the rack) contains a network switch
used to provide network connections for the servers in the rack and some slots may be left empty to allow
air flow to keep the servers from overheating.

84 inches

up to 42 servers

Figure 1.2 Illustration of a rack that holds multiple servers.

The Economic Motivation For A Centralized Data Center

• Collecting servers into a small place has an important advantage: lower cost. Specifically lowering:

o Operating expenses (opex) - recurring cost

o Capital expenses (capex): - equipment cost

• The availability of high-speed computer networks allows an organization to optimize costs by consolidating
server equipment into a single physical location (date center).

Origin Of The Term “In The Cloud”

The cloud represents a set of networks and the equipment that interconnects the networks. Saying a data center is “in
the cloud” is technically inaccurate because the servers in a data center are not part of the Internet itself.

Servers are merely computers that attach to the Internet, and should be depicted outside the cloud, with network
connections leading to the cloud.

Although technically inaccurate, the phrase in the cloud arose because early data centers were located close to
networking equipment at Internet peering points rather than in separate buildings.
Centralization Once Again

• For decades, the low cost of computers encouraged decentralization. The power wall and cost of IT
staffing favor a return to a centralized model that consolidates computing facilities into data centers.