ECEN5553 Week 3

Download Report

Transcript ECEN5553 Week 3

ECEN5553 Telecom Systems
Week #3
Read
[4a] "High Speed Ethernet: A Planning Guide"
[4b] "What If Ethernet Failed?"
[4c] "8 Ethernet Predictions for 2014"
[5a] "Is Your Ethernet Fast Enough?"
[5b] "Showdown Coming on Ethernet Standard to Serve Faster
Wi-Fi"
[6a] "Browse at Your Own Risk"
[6b] "The Data Brokers: Selling Your Personal Information"
Exam #1 Lecture 15, 21 September (Live)
No later than 28 September (Remote DL)
Outline 7 October 2015, Lecture 22 (Live)
No later than 14 October (Remote DL)
Outlines
Received
due 7 October (local)
14 October (remote)
0%
802.3 MAC Flow Chart
No
Packet to
Send?
Drop Packet.
Notify Higher Layer
Yes
Set Collision Couter
=0
Traffic on
Network?
No
Back-Off
Yes
16th Collision?
Yes
Bump Collision
Counter by +1
No
Send Packet
Yes
Collision?
No
Jam
Major Drawbacks of CSMA/CD
MMAT equals infinity
(No guaranteed Bandwidth)
 No Priorities

These make 802.3 Ethernet marginal,
at best, for Multimedia Traffic.
802.3 Packet Format
Bytes: 7
Pre
1
6
6
SFD
Destination
Address
Source
Address
46-1500
Data + Padding
2
Len
4
CRC
Preamble
Logic
One
0
Logic
Zero
0
volts
+1
Series of pulses
generated at
receiver T seconds
apart & in middle
of each symbol.
0
time
-1
T
Transmitting a File
Broken into smaller packets
 Initial packets from Layer 5
Open Logical Connection
 Packets from Layer 7
“Data” Contains Layer 7 traffic
“Data” Contains Layer 3-6 info
 Packets from Layer 4
Acknowledgements
 Final packets from Layer 5
Close Logical Connection

10Base5 & 10Base2 (Obsolete)
Coax Cable
PC
PC
Printer
Logical & Physical Bus
All nodes monitor traffic
Nodes share 10 Mbps
10Base5 "Vampire Tap"
10Base2 "T" connection
Images from Wikipedia
10BaseT & Shared Hub
PC
Twisted Pair
PC
Hub
PC
PC
Logical Bus & Physical Star
Shared hub (OSI Level 1) copies input bits to all outputs.
All nodes monitor traffic. 4 nodes share 10 Mbps.
10BaseT & Switched Hub
PC
PC
Switch
PC
PC
Logical Bus & Physical Star
Switched Hub (OSI Level 1 & 2) copies packet to proper output.
Only the destination monitors traffic.
This example system can move up to 20 Mbps
provided the packet source & destinations differ.
10BaseT & Switched Hub
PC
PC
Switched
Hub
PC
PC
Logical Bus & Physical Star
Each node shares 10 Mbps
with the Switched Hub.
10BaseT & Switched Hub
PC
reception is
screwed up
PC
PC
Switched
Hub
PC
Using Half Duplex 10BaseT,
a collision occurs if PC & Switched Hub
simultaneously transmit.
IEEE 802.3u
100 Mbps Fast Ethernet
Preserves CSMA/CD
 Preserves Packet Format
 Maximum End-to-End Lengths (a.k.a.
Collision Domain) reduced to keep
Normalized Propagation Delay low
 Sales exceed 10 Mbps as of ‘98

Source:
"A Roadmap to 100G Ethernet at the Enterprise Data Center"
IEEE Communications Magazine, November 2007
Ethernet Switch Port Sales
Back around 1994 to 1995…
Two 100 Mbps "Ethernets" introduced
 Version A

 CSMA/CD

Version B
 Demand

802.3u Fast Ethernet
IEEE said Version B is not Ethernet
 IEEE

Priority MAC, Ethernet Frames
IEEE said Version A is Ethernet
 IEEE

MAC, Ethernet Frames
802.12 100VG-AnyLAN
802.12 is currently Dead
IEEE 802.3z
1 Gbps Ethernet (1998)
Uses an extended version of CSMA/CD,
including "Frame Bursting"
 Best performance uses full duplex
connections & switched hubs

 CSMA/CD
included so it can be called Ethernet
Collision Domain same as Fast Ethernet
 Preserves Packet Format
 Fiber or Cat6 Cables

Full Duplex System
PC
PC
Switched
Hub
PC
PC
All > 10 Gbps, most 1 Gbps, & many 100 Mbps systems are Full Duplex.
Net IC’s are designed to simultaneously transmit & receive.
Line no longer shared. No Collisions. No need for CSMA/CD.
IEEE 802.3ae
10 Gbps “Ethernet” (2002)
Standard as of June 2002
 Does not use CSMA/CD

 Uses
switched hubs & full duplex
connections
 Uses Ethernet frame format

Initial available products used fiber
 Copper
cabling now an option
IEEE 802.3ba
40 & 100 Gbps “Ethernet” (2010)
Standard as of June 2010
 Does not use CSMA/CD

 Uses
switched hubs & full duplex
connections
 Uses Ethernet frame format

Copper cabling an option
 7-10

m, 10 twisted pairs required
Mostly uses fiber
IEEE 802.1p Priority Tags
8 priorities
 MAC protocols remain unchanged
 Used by 802.1p enabled switches

 Allows
interactive voice or video to receive
preferential treatment on an Ethernet LAN
Many Ethernet Physical Layer Standards Exist
source: "Evolution of Ethernet Standards in the IEEE 802.3 Working Group", IEEE Communications Magazine, August 2013
LAN Throughput




Average bit transmission rate actually available for use
Throughput = Efficiency * Line Speed
Shared Half Duplex Network
 No Load Efficiency: ≈ 100%
These are
 Actually a little less since frames
Estimates.
can't be transmitted back-to-back
 Heavy Load Efficiency: ≈ 1/(1+5*NPD)
 Shared network: apply to entire network
 Switched network: apply between switched hub
& end device
Switched Full Duplex Network Efficiency
 ≈ 100%
10BaseT & Shared Hub
53 m
PC
8m
PC
Hub
PC
26 m
17 m
PC
Logical Bus & Physical Star
Shared hub (OSI Level 1) copies input bits to all outputs.
All nodes monitor traffic. 4 nodes share 10 Mbps.
Max end-to-end distance is 79 meters.
10BaseT & Switched Hub
53 m
PC
8m
PC
Switch
PC
26 m
17 m
PC
Logical Bus & Physical Star
Switched hub (OSI Level 1 & 2) is packet & MAC aware.
Nodes don't see all traffic.
Line shared between node & switch.
η Distance to use is PC to Hub specific.
Ethernet Performance

Simulations show CSMA/CD is very
efficient for slow speed Networks.
 Shared
Ethernet efficiency equation reasonably
accurate.

Simulations also show that Average Delay to
move a packet at head of queue is usually
small, even under heavy load conditions.
Shared 802.3 LAN Efficiencies
500 m end-to-end
Heavy Load Conditions
 100 B packets: Formula = .8809 efficiency
5 nodes:
.9986 simulated
50 nodes:
.6980 simulated
 1500 B packets: Formula = .9911 efficiency
5 nodes:
.9953 simulated
50 nodes:
.9532 simulated
 Conclusion:
Heavy Load η reasonably accurate

Head of Line Performance
185 m end-to-end, 130 byte packets
 5 nodes
> 90% of packets do not collide
Average collisions/packet = .05
Average delay to transmit = 51 microsec.
Maximum delay to transmit = 11.3 msec.
 50 nodes
> 45% collide one or more times (2% 16x)
Average collisions/packet = .93
Average delay to transmit = 340 microsec.
Maximum delay to transmit = 236 msec

LAN History
IEEE 802.5 Token Ring







Based on early 1980’s technology
Covers OSI Layers 1 & 2
4 or 16 Mbps Line Speed
Logical Ring
A ‘Token’ is passed around the ring
Node must have the Token to transmit
Guaranteed Bandwidth
Has Priorities
IEEE 802.5 Token Ring


Technically Superior to shared Ethernet
Similar evolution to Ethernet
 Logical
& Physical Ring
 Logical Ring, Shared Physical Star
 Logical Ring, Switched Physical Star

100 Mbps products available in ’98
3

years after Fast Ethernet
Sales have crashed. 802.5 is dead.
Ethernet & Token Ring
Shared Network Efficiency
1.0
Efficiencies
Token Ring 1/(1 + NPD)
Ethernet 1/(1 + 5*NPD)
0.5
0.0
.01
.10
1.0
NPD
10.0
100
Shared Network Performance Issues
Slow Speed Network?
Both Ethernet & Token Ring work well
 Borderline Network?
Token Ring offers clearly superior
performance
 High Speed Network?
Both stink.
 Token Ring and Ethernet MAC’s don’t
scale well to long distances or high speeds

Shared Ethernet Efficiency
Designed to operate as "Low Speed"
1.0
Standard CSMA/CD
0.5
0.0
.01
.10
1.0
NPD
10.0
100
Low Speed Network?
This configuration...
PC
Server
Hub
PC
PC
10BaseT & Shared Hub
... is as good as this one...
PC
Server
Switched
Hub
PC
PC
10BaseT & Switched Hub


...IF traffic mostly going to/from same machine
Switched Hub better if diverse traffic flow
This configuration is even better.
PC
10 Mbps
1 Gbps
Server
Switch
100 Mbps
PC

10 Mbps
PC
Server on a higher speed line.
Shared Ethernet Efficiency
Gbps has higher NPD
1.0
0.5
0.0
.01
.10
.1
1.0
NPD
1
10.0
10
100
High Speed Network?
This configuration has
horrible throughput.
PC
Server
Shared
Hub
PC
PC
1 or 10 Gbps & Shared Hub
Under heavy load, too much time
spent recovering from collisions.
Ethernet (Shared) Hub
Operates at OSI Level 1
 ‘Electric Cable’
Traffic arriving at an input is immediately
copied to all other ports on a bit-by bit basis.
 Used on LAN's. Pretty much obsolete.
 Repeater = single input & single output hub

 Not
used much on Ethernet any more
 Generally now only used on WAN long haul
 May
be different protocol than Ethernet
Black Box Performance...
From
Node A
Node B
Node C
To
Node A
OSI Level 1
LAN Hub
Node B
Node C
Two packets simultaneously show up at input...
Black Box Performance...
From
Node A
Node B
Node C
To
Node A
OSI Level 1
LAN Hub
Node B
Node C
... will overwrite each other, i.e. garbage out.
a.k.a. Shared Hub
Black Box Performance...
OSI Level
1-2 (Switch)
or 1-3 (Router)
Two packets simultaneously show up at input...
Black Box Performance...
OSI Level
1-2 (Switch)
or 1-3 (Router)
... one will be transmitted (when allowed by MAC),
the other momentarily buffered...
Black Box Performance...
OSI Level 2/3
Switch
or Router
... and then transmitted.
10BaseT & Shared Hub
53 m
PC
8m
PC
Hub
PC
26 m
17 m
One big collision domain.
PC
10BaseT & Half Duplex Switch
53 m
PC
8m
PC
Switch
PC
26 m
17 m
PC
Four smaller collision domains.
Example
If Box 1 & 2 are Level 1 Hubs
One Big Collision Domain
15 Nodes share 10 Mbps
Each node gets average of 10/15 Mbps
World
10BaseT
7
Hub 1
Users
Hub 2
7
Users
Right Side to World gets
7/15th of available BW,
on average.
Example
If Box 1 & 2 are Level 2 Switches
Each node shares 10 Mbps with Switch
Right Hand Side is on one 10 Mbps line.
World
10BaseT
7
Users
Sw 1
Right hand side sees
increased delays.
Can be alleviated with
100 Mbps Box 1 ↔ Box 2 link.
Sw 2
7
Users
Right Side to World gets
1/8th of available BW.
_____
(Was 7/15th)
Switched Hubs or Bridge
On Power Up know nothing
 When a packet arrives at an input port...

 Look-Up
Table consulted
 Source MAC address not in table?
 Table
Updated: MAC address & Port matched
 Destination
 Packet
broadcast to all outputs (a.k.a. flooding)
 Destination
 Packet

MAC address not in table?
MAC address in table?
shipped to proper output
Look-up Table update is dependent on
packet arrivals
Router
Operates at OSI Layers 1, 2, & 3
 Capable of making complex routing decisions

 ‘peers
into’ packets and examines Layer 3 address
Very useful on Large Networks with multiple
end-to-end paths
 Routers frequently exchange connectivity info
with neighboring Routers

 Routing
Algorithms used to update
Routing (Look-Up) Tables
 Tables updated independently of traffic
Bridging versus Routing

Ethernet Bridge, Switch, or Switched Hub
 Uses
Layer 2 MAC Address
 Unknown Destination? Flooded
 Look-up Table updates are packet dependent

Router
 Uses
Layer 3 Internet Protocol Address
 Unknown Destination? Default location
 Look-up Tables updated independently of traffic
Small Network? Doesn't matter
 Big Network? Floods not a good idea.

Ethernet Switch
Uses MAC Source Address to populate Look-Up Table
Uses MAC Destination Address & Table for I/O Decision
Bytes: 7
1
6
6
2
Pre
SFD
Destination
Address
Source
Address
46-1500
Data + Padding
Len
4
CRC
Router
Populates Look-Up table independently of traffic.
Uses Destination IP Address & Table for I/O Decision
Bytes: 7
1
6
6
2
MAC
Destination
Address
MAC
Source
Address
20
20
6-1460
4
IPv4
TCP
Data + Padding
CRC