Transcript Chapter05 - YSU Computer Science & Information Systems

Chapter 5
Hardware and Software
Trends
Introduction
Four key areas have fueled the
advances in telecommunications
and computing
Semiconductor fabrication
Magnetic recording
Networking and communications
systems
Software development
Exponential Growth
Gordon Moore (a founder of Intel)
observed a trend in semiconductor
growth in 1965 that has held firm for
close to 40 years
Moore’s Law states that the number of
transistors on an integrated circuit
doubles every 18 months
Similar performance curves exist in the
telecommunication and magnetic
recording industries
Semiconductor Technology
 The transistor was invented at Bell Labs in
1947 by John Bardeen, Walter Brattain, and
William Shockley
 Semiconductors form the foundation upon
which much of the modern information
industry is based
 Advances in process have allowed system
designers to pack more performance into more
devices at decreased cost
Trends in Semiconductor
Technology
1.
2.
3.
4.
5.
6.
Diminishing device size
Increasing density of devices on chips
Faster switching speeds
Expanded function per chip
Increased reliability
Rapidly declining unit cost
Semiconductor
Performance
 Electricity (electrons) moves at speeds close
to the speed of light (186k miles/sec)
 As switching elements of a semiconductor get
smaller, they can be placed physically closer
together
 Since the absolute distance between elements
shrinks, device speed increases
 Semiconductor manufacturing cost is more
related to number of chips produced rather
than number of devices per chip
Semiconductor
Performance
As device size shrinks, performance
improves and capability increases
(more logic elements in the same size
package and those elements operate
faster)
During the period from 1960 to 1990
density grew by 7 orders of magnitude
3 circuits to 3 million
By 2020, chips will hold between 1 to 10
billion circuits
Semiconductor Processes
Semiconductors are produced in
processing plants called fabs
Fabs produce semiconductors on silicon
wafers
The wafers are sliced from extremely pure
silicon ingots and polished
These wafers can range in size from 6 to 12
inches (150 to 300 mm) in diameter
Newer fabs process larger wafers
Semiconductor Processes
Current state of the art fabs
process 300 mm wafers
It costs $1.7 billion dollars and
takes 30 months to construct and
equip a fab
Fabs are completely obsolete, on
average, in seven years
Semiconductor Processes
Each wafer holds many identical copies
of the semiconductor
The wafer moves from process to
process across the fab, slowly being
built up to create the final product
The last step in the process slices the
wafer up into the individual chips which
are tested and packaged
Semiconductor Processes
From early in the design of a fab, the
number of wafers the plant can process
per month is determined
To maximize return on capital
investment, the process engineers
attempt to produce the greatest
number of the highest value chips
Decreasing device size increases both
the number of chips per wafer and the
speed of the devices produced
Semiconductor Processes
The drive to use larger wafers
stems from the economies of scale
2.5 times as many chips can be cut
from a 300 mm wafer as compared to
a 200 mm wafer
300 mm fabs cost 1.7 times as much
as 200 mm ones
Device Geometries
Device geometry is defined by
minimum feature size
This is the smallest individual feature
created on the device (line, transistor
gate, etc.)
Current feature size in leading edge
fabs is 0.10 microns
Human hairs are 80 microns in
diameter
QuickTime™ and a
Sorenson Video 3 decompressor
are needed to see this picture.
Roadblocks to Device
Shrinkage
Most common chips are made using the
Complementary Metal Oxide
Semiconductor (CMOS) process
Chips using CMOS only consume power
when logic states change from 1 to 0 or
0 to 1
As clock speeds increase the number of
logical operations increases
Roadblocks to Device
Shrinkage
As the minimum feature size decreases,
components are closer together and the
number of components per unit area
increases
Both these factors increase the amount
of waste heat needed to be removed
from a device
Effectively removing this heat is a big
challenge
Industry Success
Success of the semiconductor
industry is driven by huge budgets
for scientific research, process
design, and innovation
Since the semiconductor was
invented, the industry has
experienced a growth rate of 100
times per decade
Industry Innovation
Increases in device processing power
comes not only from increased clock
rates and decreased device sizes
Innovation in physical computer
architecture also drives performance
Bus widths have increased from 8 to 16 to
32 and now are growing to 64-bit wide
With wider busses, more data can be
transferred from place to place on the chip
simultaneously, increasing performance
Industry Innovation
Cache Memory – Fast, high speed
memory used to buffer program data
near the processor to avoid data access
delays
Super scalar designs – designs that
allow more than one instruction to be
executed at a time
Hyperthreading – adding a small
amount of extra on-chip hardware that
allows one processor to efficiently act
as two, boosting performance by 25 %
Semiconductor Content
Microprocessors comprise less than
50% of total chip production
Memory, application-specific integrated
circuits (ASICs), and custom silicon
make up the bulk of production
The telecommunications industry is a
huge driver worldwide as cell phone
penetration increases
Summary
The invention and innovation of the
semiconductor industry has been
enormously important
Chip densities will continue to increase
due to innovation in physics,
metallurgy, chemistry, and
manufacturing tools and processes
Semiconductors will continue to be
cheaper, faster, and more capable
Recording Technologies
As dramatic as the progress in
semiconductor development is,
progress in recording technologies is
even more rapid
Disk-based magnetic storage grew at a
compounded rate of 25% through the
1980s but then accelerated to 60% in
the early 1990s and further increased
to in excess of 100% by the turn of the
century
Exploding Demand
As personal computers have grown in
computing power, storage demands
have also accelerated
Operating systems and common application
suites consume several gigabytes of storage
to start with
The World Wide Web requires vast amounts
of online storage of information
Disk storage is being integrated into
consumer electronics
Recording Economics
At current rates of growth, disk
capacities are doubling every six
months
Growth rates are exceeding Moore’s
Law kinetics by a factor of three
Price per megabyte has declined from 4
cents in 1998 to 0.07 cent in 2002
Bit Density
Data density for disk drives is measured
in bits per square inch called areal
density
Current areal density is 70 gigabits per
square inch and is expected to climb to 100
gigabits per square inch by the end of 2003
By 2007, areal densities are expected to
exceed 1000 gigabits per square inch
Hard Drive Anatomy
Data is stored on hard drives in
concentric circles called “Tracks”
Each track is divided into segments
called “Sectors”
A drive may contain multiple disks
called “Platters”
Writing or reading data is done by small
recording heads supported by a mobile
arm
Hard Drive Performance
Drive performance is commonly
measured by how quickly data can
be retrieved and written
Two common measures are used
Seek Time
Rotational Delay
Hard Drive Performance
Seek Time is the amount of time it
takes the heads to move from one track
to another
This time is commonly measured in
milliseconds (ms or thousandths of a
second)
For a processor operating at 1 Ghz, 1 ms is
enough time to execute one million
instructions
Common seek times of inexpensive drives
are from 7 to 9 ms
Rotational Delay
The delay imposed by waiting for the
correct sector of data to move under
the read / write heads
Current drives spin at 7200 RPM.
Faster rotational speeds decrease rotational
delay
High end server drives spin at 15000 RPM, with
surface speeds exceeding 100 MPH
Heads float on a cushion of air 3 millionths of an
inch thick
Other Performance Issues
Data transfer interfaces are
constantly evolving to keep pace
with higher drive performance.
New standards include:
Firewire
USB 2
InfiniBand
Fault-Tolerant Storage
Data has become a strategic asset
of most businesses
Loss of data can cripple and
sometimes kill an enterprise
Fault-tolerant storage systems
have become more important as
data availability has become more
critical
RAID Storage
RAID is an acronym that stands for
Redundant Array of Inexpensive Drives
RAIDs spread data across multiple
drives to reduce the chance that the
failure of one drive would result in data
loss
RAID levels commonly range from 0 to
5 with some derivative cases
RAID Tradeoffs
Creating data redundancy creates
transactional overhead and waste
of storage capacity
RAID 1 is also known as disk
mirroring where every bit on one disk
is duplicated on the mirror
Every transaction takes two reads or two
writes, and disk space is half of capacity
RAID Tradeoffs
RAID 5 spreads data across multiple
disks and creates special errorcorrecting data
With any drive failure, the lost data can
be reconstructed from the remaining
data and the error-correcting codes
This has less redundancy than a RAID 1
system, but delivers better throughput
RAID Results
Mean time before data loss (MTBDL) is
a calculation that attempts to quantify
the reliability of a drive
A four-disk storage system without RAID
has a MTBDL of 38,600 hours or about once
every four years
A five-disk RAID 5 system of equal capacity
yields a MTBDL of 48.875 million hours
CD-ROM Storage
Five inches in diameter, capable of
holding 650 MB of data
So inexpensive, powerful, and
widespread are these disks, that many
PC manufacturers are discontinuing the
sale of 1.44 MB floppy drives in new
PCs
CD-R blanks are now costing
approximately 5 cents each
DVD Storage
DVDs or Digital Versatile Discs
Store 4.7 GB of digital data
Can be used to store video, audio,
or larger data archives
Autonomous Storage
Systems
Computers have traditionally been built
with display, compute, and storage
subsystems in close physical proximity
With widespread, high speed digital
networks, these components no longer
need to be in the same physical box
Network Attached Storage and Storage
Area Networks are storage examples of
this trend
Network Attached Storage
A logical extension of the client/server
model
NAS boxes are servers not of
applications but of storage
Data storage can be centralized so that
the disciplines of archiving, security,
availability, and restoration are handled
by computing professionals, not
desktop users
Storage Area Networks
Commonly referred to as the “network
behind the server”
Create a unified storage architecture
that supports the storage needs of
multiple servers
Server to storage links are high-speed
optical connections using network-like
protocols complete with routers and
switches
Benefits of Storage
Systems
Data throughput from a server
standpoint and from a storage
standpoint must be balanced
Fast servers with slow storage or slow
servers with fast storage do not deliver
optimal performance
Decoupling storage from computation
allows managers to scale each
independently
Computer Architecture
Computers include:
Memory
Mass storage
Logic
Peripherals
Input devices
Displays
Supercomputers
At the extreme edge of the computing
spectrum, supercomputers are clusters
of individual machines lashed together
with high-speed network connections
The 50 most powerful supercomputers
in existence today are built of no less
than 64 processors
The most powerful are composed of
close to 10,000 individual processors
Supercomputer
Performance
Current benchmarking for
supercomputers is the flop or floatingpoint operations per second
The most powerful supercomputers in
the world easily exceed 1 tera-flops
The most powerful machine can attain
35 Tflops
Supercomputer Challenges
Effectively harnessing thousands
of CPUs together is a very complex
programming challenge
Massively parallel computing
operating systems are difficult to
design, optimize, and troubleshoot
Microcomputers
The first microcomputer was sold by
IBM in the early 1970s
With the progress of Moore's Law, PCs
have become more and more powerful
with desktop systems able to deliver in
excess of 2500 MIPS (millions of
instructions per second)
10000 MIPS systems will be
commonplace by the end of the decade
Trends in Systems
Architecture
Slowly systems are shifting from
being PC focused to network
focused
Client/Server Computing
With powerful graphical workstations
and high-speed networking, PCs have
become the user interface engine, not
the application
The most obvious example is the Web
browser. Any number of servers using
numerous different server programs are
all accessible by the same Web client
Thin Clients
With the “hollowing out of the
computer”, client PCs no longer
need to “do it all”
Storage can be offloaded to SANs or
NAS arrays
Compute cycles can be located on
application servers across or even
external to the enterprise
Communications
Technology
The same semiconductor and switching
technologies that have driven the
computer revolution have driven the
telecommunications revolution
Fiber-optic data capacity has increased
even faster than Moore’s Law rates for
semiconductors
Fiber-optic capacity doubles every six
months
Intranets, Extranets, and the
WWW
Intranet – Network dedicated to
internal corporate use
Extranet – Network used to bring
partners external to the company
into the corporate network
The World Wide Web
Invented by Tim Berners-Lee at CERN
Open standard client/server interface
Uses open standard HTML for page
formatting and display
The Web creates a powerful open
access structure that everyone can
leverage for business needs
WWW and Business
Intranets, extranets, and the
Internet all play parts in creating
an e-enabled business
Client/server architectures
modularize components allowing
special purpose or custom built
systems for online business
Thin Clients
Called “thin” because they have
minimal local storage, and function
primarily as display devices
Applications are executed locally
but reside remotely
Benefits of Thin Clients
Thin clients allow businesses to have a
high degree of control over user’s
desktops
Central client management eases
troubleshooting and allows rollout of
application upgrades without much
overhead
Thin clients commonly lack removable
storage so data security is enhanced
Programming Technology
As opposed to the exponential rate
of growth with the previously
discussed technologies, software
has grown at a linear pace
Operating Systems
Current examples are
Microsoft Windows XP
Linux (Open source)
Apple OS X
Free BSD (Open source)
Solaris (Sun)
AIX (IBM)
History of Operating
Systems
First programs were called “Monitors”
They allowed operators to more easily load
programs and retrieve output
Uniprocessing – executing one program
at a time
Multiprocessing – appearing to execute
several programs simultaneously by
processing a few instructions from each
in succession
Network Operating
Systems
Operating systems that
incorporate network aware hooks
so that systems can utilize
resources seamlessly across the
network infrastructure
Microsoft’s Windows 2000 and
Linux both incorporate these
elements directly out of the box
Application Programming
Internet technology requires new
tools to exploit its full potential
Markup languages such as SGML,
HTML, and XML
Java is used to code applications that
can run on a broad range of operating
systems and microprocessors