Major introductory network communication concepts

Protocol: “format and ordering of messages exchanged between two or more communicating entities…”

(Return to this next week when discussing layered nework architecture)

Host, client, server

What is client? (requests services)

What is server? (provides services)

What are examples of common client/server applications?

(www, email, telnet, newsgroups, ftp,…)

Connection-oriented service

Analogy to telephone service:

- protocol to establish / close connection (handshake)

- reliable service (error-free and correctly sequenced delivery)

Going beyond the telephone analogy:

- service may also provide flow control - keep faster one from overwhelming slower one

- (all operating systems provide this between components. How?)

- service may also provide congestion control – prevent global gridlock

- (related to but distinguished from flow control. Flow is local, congestion is global.)

- Analogy to streets of Springfield?

Internet connection-oriented service provided by TCP (Transmission Control Protocol)

Connectionless service

Analogy to snail mail service:

- message broken into packets

- each packet sent separately, maybe by different routes

- NO reliable service

o packet may not be delivered

o packet may be delivered w/ errors

o packets may not arrive in sequence

Internet connectionless service provided by UDP (User Datagram Protocol)

Connection-oriented or Connectionless service is end-to-end transport service and focus is on hosts.

Transport service hides the underlying network infrastructure ("core" in author's terminology).

Network infrastructure consists of interconnected routers.

Data must be switched through routers to make its way from sending host to receiving host.

Three basic approaches toward implementing switched communication:

1. Circuit switching : reserved and dedicated connection between hosts.

2. Packet switching : message partitioned into packets, switched separately w/o reservations

3. Message switching : packet switching where one packet contains whole message

Preliminary concept: bandwidth

Transmission capacity of given medium (or channel therein) is called bandwidth
Literally means "width of frequency band"
example: analog telephone channel (line) has 4 kHz bandwidth (e.g. from 0 Hz to 4000 Hz)
bandwidth determines maximum data transmission rate in bps (details late in the semester)

More on Circuit Switching:

Not feasible to dedicate entire bandwidth of medium to one circuit all the time. Medium is typically shared using either:

1. Frequency Division Multiplexing (FDM), or

2. Time Division Multiplexing (TDM)

(note: fiber optics gives rise to a third method, Wavelength Division Multiplexing)

Put simply:

An FDM circuit gets part of the medium bandwidth all the time,

A TDM circuit gets all the medium bandwidth part of the time.

Which type is an analog telephone line?

In FDM, total bandwidth is partitioned into equal width bands (channels).

Analog phone line channels 0Hz - 4000Hz, 4000Hz - 8000Hz, etc.

Of the channel, 3000Hz is usable with 500Hz "guard band" around it.

In TDM, time is partitioned into fixed length frames and each frame divided into fixed number of time slots.

Example: T1

- 1.544Mbps,

- 8000 frames per second,

- each frame is 193 bits

- each time slot is 8 bits

- one frame thus supports 24 channels plus 1 timing bit. 8*24+1=193

- effective throughput is 1.536Mbps (subtracting the 8000 timing bits per second).

- each channel gets 8 bits * 8000 frames per second, or 64000 bps.

The major disadvantage of circuit switching is:

since channel is reserved, bandwidth is wasted while inactive.

More on Packet Switching

Multiplexing occurs here too, only without reserved slots.

Since packets are sent randomly from various sources, this is called Statistical Multiplexing

Buffering occurs in router:

input buffer contains incoming packets
output buffer contains outgoing packets
entire packet must be received before outputting, this is store-and-forward

Sources of delay in packet switched network:

store-and-forward occurs at router
(first arriving bit cannot be transmitted until last bit arrives)
depends on transmission speed and message/packet length (a.k.a. transmission delay)
processing occurs at router
(time required for router to make routing decision)
queuing occurs at router
(packet waiting in output buffer for transmit line to become free)
propagation occurs on links
(time required for bit to travel from one router to next)
depends on speed of energy through medium and length of link.

Example:

1000 byte packet arrives on 10Mbps transmission link. What is the store-and-forward delay? If the first bit arrives at time t and the last bit arrives at time t+Dt, then Dt is the store-and-forward delay. For this example, the amount of time to transmit 1000 bytes on a 10Mbps link. In general, packet size divided by transmission rate. 8000 / 10,000,000 = .8 milliseconds or 800 microseconds.

Example:

Assuming fiber optic medium, signal propagates at near the speed of light. Assume 3 x 10⁸ meters/second, or 300,000 km/second. What is propagation delay over 3 km link?

3 km / 300,000 km per second = 1 / 100,000 second = 10 microseconds.

Example:

Assuming satellite in geosynchronous orbit (~36,000 km altitude), and unguided transmission at speed of light (300,000 km per second), what is propagation from earthlink to satellite?

36,000 / 300,000 = 120 milliseconds.

More on Message Switching

Simply a special case of packet switching: message transmitted as one packet

Which type of delay is likely to dominate in lightly-loaded earthbound network?

Compare packet with message switching with an example:

- message length is 8000 bytes (64000 bits)

- There are 2 routers (R1,R2) between the 2 hosts (H1, H2).

- each link has 1 msec. Propagation delay

- For packet switching, packets are 1000 bits

- transmission rate 1 Mbps

- assume no processing or queuing delays and infinite router buffers

Packet switching:

Total of 64 packets are required (ignoring overhead bits)

First bit of packet 1 arrives at R1 after 1 msec.

Last bit of packet 1 arrives at R1 1000/1000000 sec later; at time 2 msec.

First bit of packet 1 leaves R1 at time 2 msec.

First bit of packet 1 arrives at R2 1 msec. later; at time 3 msec.

Last bit of packet 1 arrives at R2 1000/1000000 later; at time 4 msec.

First bit of packet 1 leaves R2 at time 4 msec.

First bit of packet 1 arrives at H2 1 msec later; at time 5 msec.

- There are no queuing delays; packets are received one after the other.

- H2 gets continuous stream of bits beginning at time 5 msec.
(since all links are same xmit rate, first bit of packet P arrives at router just as first bit of packet P-1 is leaving; last bit of packet P arrives at router just as last bit of packet P-1 is leaving; at next tick the first bit of packet P leaves router – thus continuous stream)

- Since there are 64000 bits arriving at 1000000 bits per second, the final bit will arrive 64,000/1,000,000, or 64 milliseconds, later.

- Total time is thus 5 + 64 = 69 milliseconds.

Message switching

First bit of message arrives at R1 after 1 msec.

Last bit of message arrives at R1 64000/1000000 later; at time 65 msec.

First bit of message leavesR1 at time 65 msec.

First bit of message arrives at R2 1 ms. later; at time 66 msec.

Last bit of message arrives at R2 64000/1000000 later; at time 66+64 = 130 msec.

First bit of message leaves R2 at time 130 msec.

First bit of message arrives at H2 1 msec. later; at time 131 msec.

Last bit of message arrives at H2 64000/1000000 later; at time 131+64 = 195 ms.

Can estimate circuit switching by eliminating store-and-forward delay:

Assume message gets 8 bits/frame, with 8000 frames/second.

How long does it take for first bit to arrive at receiving host?

First bit arrives at receiving host after 3 msec (H1 -> R1 ->R2 -> H2)

Requires 8000 frames (64000 bit message / 8 bits per frame)

Last bit leaves sender 1 second after the first bit (8000 frames at 8000 frames per second)

Therefore last bit arrives at receiver 1.003 seconds after first bit leaves sender.

World of packet switched networks can be broken down into:

Datagram networks
Virtual circuit networks (path is established in advance during connection protocol)

Datagram:

Each packet carries destination address.
Router has routing table; each entry is (destination address, output line) pair
When packet arrives at router:
- find destination address in routing table
- transmit on corresponding output line.
The Internet uses datagrams, with IP (Internet Protocol)

Virtual Circuit:

Each packet carries VC number.
Router has VC table; each entry is (incoming line, incoming VC#, output line, output VC#)
When packet arrives at router:
- find incoming line, incoming VC# pair in VC table
- assign corresponding output VC# and transmit on corresponding output line.
Note: VC# is "local" (between 2 adjacent routers), not "global". Consider advantages of this….
ATM (Asynch Transfer Mode) uses this

Layered Network Architectures

See notes from previous semesters, specifically http://www.cs.smsu.edu/~pete/csc465/notes/spring00/osi_model.html

- OSI 7-layer model reference model

- Definition of protocol and interface in this context.

- Layer N-1 provides service to layer N.

- N-PDU (Layer N Protocol Data Unit)

- Role of headers

- Typical layer functions

Internet 5-layer protocol stack:

Application (5-PDU is message)

Transport (4-PDU is segment)

Network (3-PDU is datagram)

Link (2-PDU is frame)

Physical (1-PDU) - job is to transmit bits.

[notes | CSC 465 | Peter Sanderson | Computer Science | SMSU ]

Updated 24 January 2001

PeteSanderson@smsu.edu