I. Introduction to Networks
A.Growth of Subject
1.In the last 5-10 years
many companies have made a transition in how they organize their computer
systems. They
have moved from
distributed systems (mainframe and terminals) to networks (servers and
personal computers.)
This new approach
saves money (price/performance ratio goes up faster than linear!) and allows
greater access to
resources.
2.In the last 3 years
a.Teachers, high school students, etc. all have E-mail accounts and are
accessing information via the Internet
and the WWW.
b.WWW makes information available in an attractive format (as opposed to
Gopher, etc.)
c.Commercial networks (America Online, Compuserve, etc.) are providing
network services along with
Internet Navigators.
3.In the past, the network
was the playground of research scientists (physicists, meteorologist, etc.)
who needed
powerful computers.
4.Future
a.Communication (person-to-person)
b.Videoconferencing
c.Entertainment
B.Technical Design Issues
1.Jargon: There aren't many
equations in this field, but there are lots of specialized words and, in
particular, lots of
acronyms! Tanenbaum
has them in the appendix. They have to be learned.
2.Network types
a.Broadcast/Multicast: These send information to all (broadcast) or a subset
greater than one (multicast) of
machines on the network.
Fig. 1-1. analogy: A Professor lecturing to a class
The information that you want to send, a packet, will have a header, which
tells the machines on the
network for whom it is intended as well as data or a payload which is the
actual information to be
transmitted. It can, of course, be intended for many machines.
Note: If packets are small and all have the same size, they are cells.
(for example in ATM)
b.Point-to-Point: All connections are between individual pairs of machines,
and information hops from
machine to machine to get to its final destination.
Fig 1-2. A telephone tree is an example of a point-to-point network.
c.Generally, small-scale networks are broadcast, large-scale networks are
point-to-point.
3.Network Sizes
Fig. 1-3. Classification of interconnected processors by scale.
a.The basic sizes are LAN's, MAN's, WAN's and the Internet.
b.LAN's
These are distinguished by simpler transmission technology and topology
as well as smaller size that
cuts costs.
Bus and rings are popular topologies (both are broadcast.)
Fig. 1-4. Two broadcast networks. (a) Bus. (b) Ring.
An arbitration mechanism is needed to decide who gets to transmit. Usually,
only one machine can
transmit at a time.
Ethernet (IEEE 802.3) is on buses. (10-100 Mbps)
Token ring (IEEE 802.5) is on rings. (4-16Mbps)
c.MAN's
Essentially, these are networks based on DQDB, distributed queue dual bus
(IEEE 802.6) and also
Broadcast.
Fig. 1-5. Architecture of the DQDB metropolitan area network.
This one has two buses with uni-directional flow. There is a head end in
both directions to govern the
flow.
d.WAN's
The machines that run applications are called hosts or end systems. The
hosts are on LAN's or
MAN's that are connected through routers to a subnet.
Subnet - inter-connected routers
Net - Subnet and hosts
Fig. 1-6. Relation between hosts and the subnet.
When two messages want the same output port of a router, that is called
contention. Typically, both
packets are received, and one is immediately forwarded while the other
is stored or buffered. [This
network is point-to-point, store-and-forward or packet-switched (as opposed
to circuit-switched,
where a connection is set up in advance.)]
Managing this connection is a major network management issue and interacts
with the topology.
Note: A variety of different topologies exist and are shown in Fig. 1-7.
More connectivity implies
less contention but higher cost.
Fig. 1-7. Some possible topologies for a point-to-point subnet.
(a)Star. (b)Ring. (c)Tree. (d) Complete. (e)Intersecting rings. (f) Irregular.
When buffers overflow, packets are dropped. The users must detect this
and re-transmit.
e.Internets vs. the Internet
a.A collection of inter-connected networks is an internet. These networks
may be based on
incompatible software and/or hardware. They must be connected through a
gateway that talks to
both.
b.The Internet is a particular internet that includes virtually universities,
businesses, most US
Government organizations, etc.
C.Reference Models
1.Layering
Fig. 1-8. Layers,
protocols, and interfaces.
a.Because networks are complicated, one divides them conceptually into
layers. Each layer carries on a
conversation between hosts without knowing what is going on above or below
using a set of conventions or
rules that are called protocols.
b.Other jargon:
Corresponding layers on different machines are peers.
Below layer 1 is the physical medium through which the hosts actually connect.
The interface defines the services that a lower layer provides an upper
layer (see Tanenbaum
22-23.)
The network architecture defines the layers and protocols.
The protocol stack is the list of protocols used by the architecture.
c.Examples:
1.Tanenbaum has an example in which one philosopher wants to send another
a message "I love
rabbits" but they do not speak the same language. The philosophers exchange
messages directly with
translators (interface) who don't understand the philosophical meaning
of the message but know
enough to translate it into a language that all translators know-Dutch
in this example. The translators
directly exchange messages with the secretaries (interface) who may not
know Dutch but can
transcribe and send words (service) and then send the message (physical
medium.)
Fig. 1-9. The philosopher-translator-secretary architecture
2.Two programs, WordPerfect and Microsoft Word want to transfer a document
"Network Studies,"
preserving formatting. These are converted into RTF (rich-text-format)
which is pure ASCII and
contains the additional information that way. These are then converted
into the bit string appropriate
for the computer it is on, e.g. a PC vs. a Macintosh and then sent via
a modem.
d.At each level, headers and/or trailers are added to communicate needed
information; also packets may
need to be subdivided. At a lower layer, the layer looks at the headers
from higher levels as part of the
message! It makes no distinction between the headers and messages at higher
levels.
Fig. 1-10. Example information flow supporting virtual communication in
layer 5.
e.Each layer may manage error control sequencing and communication.
Circuits are not perfect; errors should be detected and corrected. In older
protocols (X.25 and
stacks) meant for error-prone media(copper wires), error-correction is
done at lower layers and
frequently. Also correction as opposed to just detection is mandatory.
In new protocols (ATM)
meant for relatively error-free media (optical fibers), error control can
be done at higher layers.
Tradeoff: Reducing congestion by allowing rerouting versus the additional
complexity and cost of
sorting out the packets on the end and time sensitive. [Voice and video
are time-sensitive; computer
data usually is not. Computer data is bursty; voice and video usually are
not.]
Fig. 1-11. Note: tradeoff is cost per application bit. The total number
of other bits sent should be
minimized.
Do packets arrive in the order that they are sent? If not, they must be
sorted out.
Is communication one-way or two-way? Is it connection-oriented or connectionless?
2.Connection-oriented and
connectionless
a.Connection-oriented: This is like the phone system. You establish a connection
prior to sending any
messages. The messages are then received in the order they are sent. [Messages
correspond to the
transmitted voice which cannot be garbled.] The connection is then released.
b.Connectionless: This is like the postal system. Two letters are routed
independently and a sequence of
letters does not necessarily arrive from sender to destination that it
is sent.
c.Quality of Service (QoS): Some services are reliable, implying that they
never lose data. Reliability is
usually important for data (but not electronic junk mail) but not for voice.
Fig. 1-12.
Different layers can and do have different QoS. For example: TCP/IP which
is used on the Internet offers
unreliable, connectionless service on the (lower) IP level and reliable,
connection-oriented service on the
(upper) TCP level.
3.OSI Reference Model
a.This model is based on a proposal for international standardization of
network protocols:
ISO = Internation Standards Organization
OSI = Open Systems Interconnection
It has seven layers as shown in Fig. 1-13 of Tanenbaum. In practice, layers
have different sizes; they can be
fat or thin.
b.Layer Structure
Fig. 1-13. The OSI reference model.
Physical Layer: concerned with mechanical, electrical, and procedural interfacing
with the physical
transmission medium.
Data Link Layer: Carves data into frames (typically 100-1000 bytes), transmits
them sequentially
and then re-transmits them if there are errors. There can be a duplicate
if an acknowledgement frame
is lost. This layer deals with damaged, lost, and duplicate frames. It
also deals with flow regulation so
that a fast transmitter doesn't drown a slow receiver.
MAC Sublayer: Broadcast networks must also control access to the shared
channel. This issue is
dealt with in a special sublayer.
Network Layer: This layer is aware of the whole subnet topology and is
responsible for routing
packets. Depending on the protocols, this can be static or dynamic. It
also must control congestion
in the network when too many users wish to use the network.
Transport Layer: This layer takes data from the upper (applications) layers
and packages it for
transport on the network. It can use several network connections for one
message if speed is
required or bundle several messages on one connection to reduce costs.
It sets the quality of service
[e.g., error-free point-to-point, sequential]. It also monitors data flow
from host-to-host (like
router-to-router control in the data link layer.)
Session and Presentation Layers: These two layers establish sessions and
worry about synchronizing
transfer (putting checkpoints in the data stream) and syntax (ASCII vs.
Unicode.) This layer is thin in
practice since the applications typically reach into the transport layer.
Tannenbaum generally
subsumes them under applications and so shall we.
Applications: E-mail (communication), Telnet (file transfer), Terminal
emulators (for making the legs
on a Macintosh or a PC work remotely on a SUN), etc.
c.Each layer adds its information in headers or trailers that are treated
as data by the layers below and are
stripped off as the data goes back up the chain in the receiving host.
Fig. 1-14. An example of how the OSI model is used. Some of the headers
may be null.
4.TCP/IP Reference Model
a.The Internet is based on TCP/IP; so, it is a real-world, very important
protocol. [The OSI model is very
important conceptually and more complete, but there is not existing network
that corresponds to it exacty.
b.A goal of TCP/IP is to provide connections that are maintained when routers
between the source and
destination go down.
c.Layers
Fig. 1-15. The TCP/IP reference model.
1.Host-to-Host Network: This layer, which corresponds to the data link
and physical layers is not
defined within TCP/IP and varies from network to network.
2.Internet Layer: This layer injects packets into the network and routes
them to their destinations. The
service is connectionless. Each packet travels independently to its destination.
Packets can arrive out
of order and by different paths. The protocol used at this layer is IP
(Internet protocol). Routing and
avoiding congestion are the key issues.
3.Transport Layer: This layer allows peer entities (e.g., E-mail) on the
source and destination hosts
exchange messages. There are two end-to-end protocols defined:
Fig. 1-16. Protocols and networks in the TCP/IP model initially.
TCP (transmission control protocol): It is reliable and connection oriented.
On the transmitting
end, it takes the incoming byte stream and breaks it into packets for transport.
On the
receiving end, it re-assembles them into the output sequence. It handles
flow control and error
control.
UDP (user datagram protocol): It is an unreliable, connectionless protocol.
It is used by
applications that want to handle their own flow control, error control,
and sequencing.
Examples: one-shot client-server request-reply queries and time-sensitive
messages (like
video and voice.)
4.Application Layer: The usual suspects appear in this layer: Telnet (virtual
terminal), PINE (E-mail),
FTP (file transfer), HTTP (fetching pages on theWWW), etc.
d.Note that the internet (network) layer is connectionless while the transport
layer running TCP is
connection-oriented. Thus, we see that being connectionless or connection-oriented
is a property of a
layer, not the network.
5.OSI vs. TCP/IP
a.The basic distinction between the OSI model and the TCP/IP model is between
theory and practice.
b.In OSI: the model came first and then the protocols. The protocols are
unwieldly and have not been
successfully implemented. By contrast, it has been a great success conceptually.
c.In TCP/IP: Just the opposite holds. Conceptually, it does not distinguish
well between layers, protocols,
and interfaces. In practice, it is widely used in the network and transport
layers, but it doesn't deal with the
data link and physical layers which must be separately implemented.
D.Networks and Network Services
1.For PC LAN's, a very commonly
used network system is Novell NetWare. Its protocol stack predates OSI
and
looks a lot
like TCP/lP.
Fig. 1-17. The Novell
NetWare reference model.
a.The physical and data link layers use standard LAN protocols like ethernet
or token ring.
b.At the network layer, there is an unreliable connectionless protocol
called IPX. Above that in the transport
layer, there is a connection-oriented transport protocol NCP (network core
protocol) and another SPX for
just transport.
c.Just like in TCP/IP, it is based on an internet datagram packet which
is created at the network layer. Its
structure is shown below and in Tannenbaum.
Fig. 1-18. A Novell NetWare IPX packet.
d.When a new machines boots up, it exchanges messages with routers and
servers so that they become
aware of the machine's presence and the machine becomes aware of the services
that are offered.
2.SMDS (Switched Multimegabit
Data Service)
a.This is a service offered by RBOC's (Regional Bell Operating Companies),
other regional carriers, and
long-distance carriers. It is designed to deal with bursty traffic (typical
for data) as opposed to steady traffic
(typical for voice).
Fig. 1-19. (a) Four LANs interconnected with leased lines. (b) Interconnection
using SMDS.
b.Shown in the figure is a case where it might be useful. A company wants
to hook up four geometrically
remore LAN's so that they can transfer data. They can lease lines, but
that is inefficient if they are only using
the lines a small part of the time.
SMDS transmits packets as they arrive. The transmission is connectionless
and unreliable.
To regulate the traffic, a counter increments regularly. When a new packet
is sent in, the number of bytes in
it is counted. If the number is less than the counter, it is transmitted,
it is sent at the full network speed.
Otherwise, it is dropped.
3.X.25
a.This approach, widely used outside the US was developed in the 1970's
by the CC ITT (Comité
Consaltatif International Télégraphique et Téléphonique).
b.Standards
Physical layer: X.21 specifies digital links. Actually, virtually all systems
that use X.25 have another
physical setup that is analog, based on the RV-232 standard.
Data link: A number of (slightly incompatible) variations exist. It takes
care of transmission errors on
the telephone lines.
At the network layer, it is connection-oriented and reliable; its packets
are 128 bytes. A 'personal'
circuit is set up that is called a permanent virtual circuit.
c.They operate up to 64 Kbps so they are obsolete for many purposes.
4.Frame relay: This is a
'bare bones' service that is analogous to a virtual leased line. Like SMDS,
it uses counters
to make sure
that the averaged (leased) transmission rate is not violated. Its base
speed 1.5 Mbps is much higher
than X.25. It
takes advantage of reliable digital lines by eliminating most error control
and all flow control. That
lowers cost
but requires users to provide a functionality themselves.
5.B-ISDN and ATM
a.The phone companies use a plethora of networks for different services.
POTS: plain old telephone service is the circuit-switched network used
for voice SMDS; frame relay
are packet-switched and separate.
SSN7: phone companys' signalling network, and it also separate.
The phone companies would also like to offer cable TV service.
b.A perceived solution is to invent a single digital network that can deliver
all these services as well as other
services like movies and music on demand, interactive libraries to the
home, videoconferencing, etc.
This service is called B-ISDN (Broadband integrated services digital network.)
c.The underlying technology for B-ISDN is ATM (Asynchronous transfer mode).
It is called asynchronous
because its basic packets (called cells) are not tied to a master clock.
Each cell is 53 bytes long with 5
bytes of header and 48 bytes of data.
Fig. 1-20. An ATM cell.
d.Why cell switching?
This is a giant break from circuit-switching which is the historical switching
technology in the phone
system.
Small cells imply that time sensitivity can be respected where needed.
At the same time,
broadcasting and data flow (with a significant overhead) are possible.
e.ATM is connection-oriented. While delivery is not reliable, sequence
is. (Reliability can be assured since it
is required for some applications.)[One of the complaints about ATM is
that it looks a lot like voice.]
f.ATM has its own reference model (why not?) which is not the same as the
OSI model.
Fig. 1-21. The ATM layers and sublayers, and their functions.
The physical layer deals with the physical medium and "transport convergence."
In ATM jargon,
interfaces are replaced by convergence sublayers.
The ATM layer defines the header fields of the cells and also deals with
establishing and releasing
virtual circuits and with congestion.
The AAL layer deals with breaking up the application data into cells and
re-assembling them.
Different AAL's are needed for different data.
Fig. 1-22. The B-ISDN ATM reference model.
Fig. 1-23.
g.The ATM reference model is actually 3-dimensional. The user plane deals
with data transport, flow control,
and error control. The control plane deals with connection management.
Then, the layers and planes must
be managed.
h.The layers have sublayers and the top-most and bottom-most sublayers
deal with getting a host of different
applications with very different needs to talk to the same protocols and
getting these protocols to talk to a
host of different transmission media.
i.My Own Perspectives (see Tannenbaum for his)
ATM was designed by committee (like OSI) and is a complicated mess for
that reason.
ATM will come as a high-speed transport mechanism, but I am very unsure
that B-ISDN will come
and take over all services.
E.Standardization
1.In order to assure some
ability to communicate between different computers, phones, networks, etc.
Standards
have to be established.
2.Increasingly, vendors
are building to "open" standards, rather than using "closed" standards
(e.g., UNIX vs.
VAX-VMS) because
buyers have repeatedly demonstrated that they will bug inferior "open"
products (arguably,
VMS v. UNIX
in operating systems, Apple vs. DOS machines in PC's, etc.)
3.Standards can be:
defacto: standards
that have just happened (e.g., UNIX or the IBM PC).
dejure: have
been established by one of the standards bodies.
4.Telephone Standards
a.In the US, one must mediate between the long-distance carriers (AT&T,
MCI, Sprint, etc.) , the RBOC's
(Bell Atlantic, Pacific Bell, etc.), and other local carriers (1500, including
GTE).
b.Internationally, most countries have a national phone system, although
the worldwide trend is toward more
privatization.
c.The ITU (Internation Telecommunication Union) sets the standards for
communications. Since 1947, it is
an agency of the United Nations. It has three sectors:
ITU-R: Radio communications
ITU-T: Telecommunication standards
ITU-D: Development
Our primary concern is the ITU-T which is the successor organization to
the CC ITT (as of 1993).
d.Working groups in ITU make recommendations for phone systems (and networks)
that are widely
respected and all major phone and telecommunications companies are included.
5.There is also ISO (International
Standards Organization), a voluntary, non-treaty organization whose members
include most
of the nation standards organizations.
ANSI: American
National Standards Institute
BSI, AFNOR, and
DIN are corresponding organizations in Britain, France, and Germany.
ISO has 200 technical committees; TC97 deals with computers and information
processing.
Each TC forms working groups; the working groups have over 100,000 members.
6.Other important sources
for standards:
NIST: National Institute of Standards and Technology
IEEE: Institute for Electrical and Electronics Engineers
7.The Internet has its own
standardization procedures which are described by Tanenbaum.
|
