Transcript Tuesday, February 7, 2007 (Intro to the Network

15-441
Communications and Networking
Lecture 7
Gregory Kesden
The switching portion of these slides evolved from Prof. Steenkiste’s slides circa 2000.
Real-World Link Layer Protocols:
“Original” Ethernet








Wired physical layer
1-persistence CSMA/CD
Manchester encoding
Binary exponential backoff upon collision
IEEE 802.3, the adopted standard, has a slightly different frame format
than the original Ethernet – but the distinction is not drawn in
conversation.
10Base5, “garden hose” wiring (now obsolete) – up to 500m and 100
stations
10Base2, “think coax” -- up to 185m and 30 stations
10Base-T, “twisted pair”, up to 100m and 1024 stations
Real-World Link Layer Protocols
Ethernet, cont.

Binary exponential backoff?

Upon a collision, wait and try again, up to 16 times.

How long to wait




For the 0-9th collisions, pick a random number between
[0, 2i-1], and skip that number of slots (opportunities to send).
For collisions 10-15, stick with [0, 210-1], which is 1023
Give up after that.
Real-World Link Layer Protocols
Ethernet, cont.

What exactly is a slot?







The time to send a minimum frame.
The standard sets a maximum length of 2500m
The reference configuration is a 10Mbps LAN with 4 repeaters
Given this, the maximum round-trip propagation time is
approximately 50 microseconds (10-6 seconds).
At the spec’d 10Mbps, this makes for a 500 bit minimum frame
size. Add a bit of padding for safety and round up to a power-oftwo and get a minimum frame size of 512 bits.
So the minimum frame time is 51.2 microseconds. This is a slot
time.
Obviously, this increases as the speed of the underlying network
increases, e.g. 6400 bytes for a 1Gbps network of up to 2500m in
length.
IEEE 802.15 “Bluetooth”
S
S
S
S
S
S
M
M
S
S
S
S
S
S
Piconet
S
Piconet
Scatternet
Up to 7 active slaves/piconet, range 10m. Up to 255 parked slaves.
IEEE 802.15 “Bluetooth”, cont

RF physical layer:
 2.4GHz Band
 79 1MhZ bands
 FSK, 1 bit/Hz
 Frequency hopping at 1600 hops/second




Fairly allocates channels
Reduces multipath fading
Limited security benefit
Shares same bands as 802.1 and can interfere with each other.
• Bluetooth hops aster, so it causes more damage to 802.11 than viceversa.
IEEE 802.15 “Bluetooth”, cont.

“Baseband layer” is the lower piece of the data link layer.
 Defines slots
 Defines 625 microsecond timeslots
 even for self, odd for slaves.
 259 bits of 625 are “settling time”
 366 are usable
• 126 data access code and header
• 240 bits for data
 If five slots are combined for a single frame, 2781 of 3125 bits
are available, since settling and other overhead are needed only
once.
 A link is a logical abstraction
 Synchronized Connection Oriented (SCO) for real-time data
 Asynchronous Connection-Less (ACL) for packet-switched
data
IEEE 802.15 “Bluetooth”, cont.

L2CAP is top half of data link
 Breaks packets into frames
 Multplexing and demultiplexing of packet sources (higher level
senders and receivers)
 Quality of service negotiation – for example, balancing needs of
small packet and large-packet senders.

Application/Profile layer
 Defines 13 types of applications and higher-level stacks for them
 Dial-up, fax, cordless telephony, file transfer, synchronization,
link management, service discovery, &c
Building Bigger LANs
A Bridge-based Network





Switches are connected by point-point links.
Packets are forwarded hop-by-hop by the switches
towards the destination.
– Forwarding is based on the address
How do nodes exchange packets over a link?
How does a switch work?
How do adjacent switches manage the link?
Switch
PCs at
Work
Point-Point
link
PC at
Home
Bridges

Multiple LANS may be connected to form one logical
LAN

Since bridges are at the link layer, they do not examine
network headers, &c.

Reasons
 to increase scale,
 control load,
 allow for long distances,
 ease administration,
 security/protection
LAN
Bridge
LAN
Bridge Architecture


Takes in packets in one interface and
has to forward them to an output
interface based on the address.
– A big intersection
– Same idea for bridges, switches,
routers: address look up differs
Control processor manages the switch
and executes higher level protocols.
– E.g. “which way?”, &c.

The switch fabric directs the traffic to
the right output port.

The input and output ports deal with
transmission and reception of packets.
Control
Processor
Input
Port
Output
Port
Output
Port
Input
Port
Switch
Fabric
Output
Port
Input
Port
Output
Port
Input
Port
Bridge Fabric Options
Crossbar switch.
»Requires lots of hardware but good performance
»Multistage interconnection networks an alternative
Bus-based
switches.
»Fabric consists one (or more) fast shared buses
»Each input port has a slot time slot on the bus
Shared
memory switch.
»Switch is one large memory
»Input ports write packets to memory and output ports read packets
from memory
»Does not scale well: need very fast memory
Hybrid
solutions.
I/O Port Functions



Input port identifies the outgoing port and buffers packets if there is
contention for the switch fabric.
Output port queues packets and a scheduler determines the order in
which packets are sent over the outgoing link.
Many buffering options exist.
– Input buffering, output buffering, internal buffering
– Typically a combination is used
– Buffer management can limit throughput, e.g. head of line
blocking
Address
Lookup
Switch
Fabric
Scheduler
Address
Lookup
A Simple Bus-based Architecture
Input
Ports
Bus
Output
Ports
A Crossbar Architecture
Input
Ports
Output
Ports
The Knockout Architecture
input
buses
concentrators
buffers
Transparent Bridges








Backward learning
“Plug and go”
Listen to traffic on all interfaces
Store each machine that transmits in hash table along with
interface.
Periodically purge old entries, just in case a machine
moves.
When a frame “hits the wire”, look it up in the hash table
and forward it to the correct LAN.
If it originated on the correct LAN, do not forward it.
If the destination of the frame is not in the hashtable, flood
all attached LANs.
Transparent Bridges, cont.
frame1
LAN
frame2
Bridge
LAN
host
frame12
Bridge
frame
frame22
Consider what can happen if a redundant bridge is used for
“robustness”. Both bridges could copy the frame, creating a
duplicate.Then, this duplicate could be copied twice, and so
on…forever
Bridges with Spanning Trees

The solution is for the bridges to communicate with each other and
build spanning trees that represent the network.

One bridge is selected to be the root of the tree, perhaps by serial
number “bullying” or other broadcast-based approach.

Then, a minimum spanning tree is constructed from each LAN,
through necessary bridges, to the root.

This spanning tree is used to determine how to forward a frame.
Connections that are not in the tree cannot be used – even if some
bridges are “left out”.

The algorithm continues to run to stay up-to-date and detect bridge
failures, host moves, &c.
Switched Ethernet





Ethernet switches are a type of bridge that have a many-to-one point-to-point
relationship.
Many low-bandwidth legs share a high-bandwidth common bus.
Typically one leg, has a very high bandwidth – consider it to be the uplink.
Imagine several floors or departments, each of which has its own leg, and then
one leg going out to an intranet or the Internet.
Each port on the switch forms its own collision domain. If multiple stations are
connected on the same port, as through a concentrator or hub, collisions can
occur among them.
Among ports, collisions can be mitigated by buffering inputs
1Gbps
100Mbps switched legs
The Network Layer
Application
The solution-specific protocol used by the application
program.
Presentation
Hides host-specific and/or user-specific nuiances
Session
Transport
You are here
Network
Data Link
Physical
Extends the host-to-host abstraction provided by the
Transport layer to include more specific, complex features.
Hides the network(s) from the user and provides a
host-to-host(s) abstraction.
Moves data from one network to another
Packages data for transmission/reception over a single
network.
Transmits and receives via a particular media over a
single network.
A Network Packet
Network Layer
Packet Header
Packet Payload
Packet
Link Layer
Frame Header
Frame Payload
Frame
•Much as the physical layer wasn’t concerned with the framing performed by the link layer,
the link layer just views the network layer’s headers, &c just as it does the user’s data –
as payload.
•From the perspective of the link layer, the network layer packet is just data.
The Network Layer

The physical layer is important, because we need
some way of transmitting and receiving signals
over the physical media.

The data link layer is important, because we need
some way of organizing the communication to
allow multiple hosts to send and receive messages
given the ability of the media to carry them.

The network lay is important because hosts that
are members of separate but interconnected
networks may want to communicate.
Encapsulation
Application
data
Application
data
Application
Presentation
Presentation
Presentation
Session
Session
Session
Transport
Transport
Transport
Network
Packet (data)
Network
Packet (data)
Data Link
Frame(packet)
Data Link
Frame(packet) Data Link
Physical
current
Physical
current
Network
Physical
Network vs. Internetwork
What distinguishes “separate, but interconnected networks” from a single “internetwork”?

One trivial (and non-informative) answer: If a data link protocol is
sufficient for the machines to communicate, they are part of a single
network. If a network-layer protocol is required, the machines are on
separate networks that, compose (at least in part) an inter-network.

A slightly better answer: If the machines are all connected to the same
communications channel, they are part of a single network

An even better answer: If the machines are connected to a single
communications channel, or several communications channels that
logically form one communications channel (as through bridging), they
are all part of the same network.

A yet better answer – If the answer to the question “How do I get there?”
is interesting, the machines are not connected via one network – they are
connected via an inter-network.
The Bottom Line

The job of the network layer is to answer
the famous question, “How do I get there
from here?”
Bridges “Know it All”

Earlier today we discussed connecting communication
channels together using bridges to form a single network.

The important thing to remember about bridges is that they
are designed to combine to channels to make them operate
like one.

But think about what they must do: learn the location of
every host on the network.
Bridges: Too Much To Remember

On those occasions when they don’t know a host’s location, there is a
very high price – the frame must be flooded to every machine on the
network.

If there are too many machines, it is too costly to know the location of
all of them – especially since the lookup must be very, very fast.

If the network is large, traffic is likely to be very high – this makes
flooding frames to all interfaces very, very costly.

It would be impossible for one machine to know, and rapidly access,
the location of every machine on the Internet, never mind manage it
efficiently.
Routers and Network Addresses

Routers are the network-level equivalent of bridges. They
connect networks to allow communication between/among
them.

But, unlike bridges, they do not attempt to know every
machine. Instead, they take advantage of hierarchical
addressing and only know how to get a packet to the right
network.

Once at the right network, the link-layer can deliver the
frame.
Internetworks and Hierarchies

What is needed is some more extensible
scheme for finding machines

One way of dealing with scale is to organize
things into a hierarchy.

Each level of the hierarchy can handle only
one portion of the job.
Consider a mailing address
Gregory Kesden
School of Computer Science
CMU
Pittsburgh, PA USA

The letter carrier Brazil isn’t likely to know where I live! How does
a letter get from Brazil to me?

International mail is directed to a sorting station in the U.S. This sorting station knows
how to get the letter to a sorting station in Pennsylvania.

The Pennsylvania station knows how to get the mail to Pittsburgh

The Pittsburgh station knows how to get it to CMU

Etc, etc, etc.

By breaking things down into a hierarchy, we have to make more decisions, but each
decision is more manageable.
Network Level Addressing

Data link layer addresses are “flat” not hierarchical. As discussed,
these don’t scale well and, as a result, limit the size of an individual
network.

Network layer addresses, much like postal addresses are hierarchical.

The first part of the address identifies the network. The second part of
the address identifies the individual host within the network.

The network:host pair is globally unique, but the host id, without the
network id, is not necessarily unique.

Sub-networks (“sub nets”), or hierarchies with an individual network,
are also possible.
Internet Protocol (IP) Addresses

There are three different classes of IP addresses:

Class A: Those used for large networks (typically very
large corporations. Exception: MIT has a class A
address. Some other universities had them in the past,
but voluntarily gave them up, because there are very
few – MIT didn’t give theirs up.)

Class B: Those used for medium sized networks

Class C: Those used for small networks
Address Resolution Protocol (ARP)

ARP is the protocol used to convert from an IP address to a MAC address.

A host that has an IP address, but needs to know the MAC address broadcasts a request
to all machines on the LAN. This broadcast uses the LAN’s broadcast address. Each host
on the LAN receives this request. The host associated with the request IP address
replies.

The host with the matching IP address unicasts its MAC address to the sender. The
results of the ARP request are cached on the requestor.

Failure to flush the ARP cache after moving interface cards around is a common cause
of self-solving mysteries for novice sys. admins.

“ARP Storms” can result when many systems are turned on at the same time.

ARP includes no security – requestors believe whatever they are told from whomever
they are told.

There is a “reverse ARP”, RARP, that can convert a MAC address to a name, but it is
complex and has largely been replaced by DHCP.
IP Addresses
Class A
Class B
Class C
1
7
24
0
Network
Host
2
14
16
10
Network
Host
3
21
8
110
Network
Host
IP Addresses: A Few Thoughts

There are very few addresses for large networks
and very many for small networks – this makes
sense given the number of parties needing 224 host
LANs versus those needing 28 host LANs.

There are 232 (about 4 billion) possible addresses,
but many are lost due to fragmentation – most
groups get blocks of addresses for a whole
network, but don’t use nearly so many machines.
This fragmentation has generated a shortage of
network addresses.
How Does a Host Get an IP Address?

Ethernet MAC addresses are “built in” serial numbers
provided by the manufacturer (in theory).

But this won’t work for IP addresses – their address must
be related to the network on which they live.

A network’s administrators is assigned a block of addresses
(a whole network worth) from IANA (Internet Assigned
Number Authority).

The administrator then assigns these addresses to
individual devices.
Static Address Assignment

The easiest way to assign IP addresses is a “once and for all approach: the
administrator statically assigns an IP address to each device. With luck, the
user will give it back when the device dies or is retired.

This devices uses this address every time.

Static assignment isn’t particularly efficient for some types of devices

Mobile hosts, home machines, &c may only require network connectivity
during part of the day, but will require an IP address during this time.

If an address is permanently assigned to these transient devices, it is wasted
much of the time.

Users often don’t return the address of dead or retired machines – this is also
wastage.

This is bad, since IP addresses are in short supply.
Dynamic Address Assignment and
DHCP

In order to avoid the wastage associated with permanent, static IP
address assignment, many networks use dynamic address assignment.
One such protocol is DHCP (Dynamic Host Configuration Protocol)

Each time a machine reboots, it sends out a discovery message via a
broadcast to a special IP address (255.255.255.255).

The DHCP server, which hears this message, replies with the assigned
IP address. The host then assumes this address.

Since hosts may be transient, the address is leased to machine for a
period of time. Unless the machine renews its lease, it will expire and
the address can be reassigned to another machine.

Obviously, the DHCP server needs to be assigned a block of IP
addresses. This is done by the system administrator.
More about DHCP

Broadcast messages to special addresses, such as the one used for DHCP are
not communicated by routers to other networks – if they were, only one DHCP
server could exist in the world.

As a result, if the same administration is responsible for multiple networks (or
subnetworks), the available addresses would need to be partitioned among the
DHCP servers. This isn’t particularly efficient and is an administrative burden.

To avoid this, DHCP relay servers can be used.

One DHCP server can be configured to handle multiple networks (or
subnetworks).

One relay server can be placed within each network.

The relay server listens for the DHCP discovery broadcast and relays it to the
DHCP server, which then responds directly to the requesting host with its IP
address.