1 | 44

Communication Systems3rd lecture

Chair of Communication SystemsDepartment of Applied Sciences

University of Freiburg2008

2 | 44

Communication SystemsLast lecture

Message segmentation in packet switched networks advantage over message switching

Different types of packet forwarding/routing in datagram networks (Internet, ...)

virtual circuit networks (ISDN, ATM, ...)

Discussed taxonomy of different network types (circuit and packet switching with respective subtypes)

Requirements for communication between so called end systems network layering and models: OSI vs. TCP/IP

depending on requirements of end users

3 | 44

Communication Systemscomparison of OSI and TCP/IP layers

Remember two the network layer models

4 | 44

Communication Systemslast lecture

Physical representation of digital bit streams all kinds of electromagnetic waves (electromagnetic spectrum)

encoding, decoding of data for transport over different types of media

physical parameters: frequency, wavelength, (effective) bandwidth, Nyquest formula for max. bandwidth of given medium

copper wire (single twisted pair for telephone, higher quality for Ethernet, fiber optics), “air” (mobile phones, satellite links, WLAN, ...)

guided and unguided media, propagation delay (speed of light)

5 | 44

Communication Systemslast lecture

Example for physical data transport: (A)DSL taken from the web interface of the Fritz!Box combined DSL modem/

Internet (masquerading) router (ADSL2+ standard compliant)

6 | 44

Communication Systemslast lecture

Example for physical data transport: (A)DSL data transport over two-wire copper, frequency division multiplexing

(FDM), ~31, ~255 sub bands of 4kHz within the bandwidth of 1MHz

upstream in lower frequency band, upstream in higher (why??)

maximum allowed symbol rate per Hz: 15Bit

within each sub band of 4kHz the max. possible symbol rate is measured (depends on signal / noise ratio – the yellow diagram, and other sources of signal attenuation)

thus you get a absolutely theoretical maximum of ~61kBit/s per sub band

channels are parallelized and higher level protocols (mostly ATM in Germany) are run on top

7 | 44

Communication Systemsplan for this lecture

Need of an universal service IP as layer 3 network protocol Start with looking at IP header

addresses and other header fields

Special IP addresses Structure of the IPv4 address space

8 | 44

Communication Systemsnetwork layer models

Talked of some base concepts of data communication and bit transportation

But how to do that in an ordered and general way? A structured composition of networks is needed for data

communication of very different machines and operating systems There are several of these models, the ISO/OSI layering model is

one of them ISO: International Standards Organization OSI: Open Systems Interconnect Reference model for implementation of network architectures

9 | 44

Communication Systemsuniversal service connecting networks

We introduced some different kinds of physical networks Modem, ISDN, DSL for connection of individuals

Technologies like Ethernet (over copper and fiber optics)

Wireless networks, ...

Many implementations for different networking purposes Wide Area Networks (WAN) which may span countries (e.g. DFN

for Germany or GEANT(2) for Europe) or even continents; connecting infrastructure components like routers not end user machinery

Local Area Networks (LAN) which implemented within buildings, mostly bridging only short distances with many hosts connected

In the past distinction of networks by bandwidth offered

10 | 44

Communication Systemsuniversal service connecting networks

Each type of network structure may require or be implemented with different low level protocols

11 | 44

Communication Systemsconnecting networks

DSL as an example for dedicated point-to-point connection over a two wire copper cable with a length up to 6km

Ethernet – packet orientated LAN protocol Multilink (broadcast) network with no dedicated point-to-point

connections

Different speeds over copper wire, coaxial cable and fiber optics

ATM – connection orientated LAN and WAN protocol Virtual ptp connection through virtual channels and paths

Modem, ISDN for WAN, FDDI, TokenRing, ... for LAN Wireless links on GPRS, HSCSD, UMTS, WLAN, ...

12 | 44

Communication Systemsconnecting networks cont.

Intermediate protocol with unified addressing scheme is needed

13 | 44

Communication Systemsconnecting networks

Universal protocol needed is implemented in the network layer Network layer is third layer in OSI model Physical layer (first layer) implements the real bit connection

over different media (e.g. Twisted pair or optical fiber with Ethernet or “air” for UMTS or GSM)

Data Link Layer contains higher level protocols of Ethernet, GSM, UMTS, ... not discussed in depth in this lecture

Each layer adds its header to a packet Needed for packet handling and routing

14 | 44

Communication Systemsconnecting networks

Helper protocols needed during packet processing inform senders on congestion or lost packets

map different layers addresses on each other

15 | 44

Communication Systemsfunctions of an universal service

Define unified addressing scheme which is hardware/software independent

Realize datagram delivery between networks Hosts – end systems as defined in first lecture Routers touch two or more networks, forward network-layer

datagrams between them (routers use layer 3) Routers execute routing protocols to learn how to reach

destinations Universal service protocol should be an open standard without

fees and licenses to be paid to get acceptance May implement/use helper protocols (ARP and ICMP for IP,

introduced in practical or theoretical exercises)

16 | 44

Communication Systemsissues of an universal service

Network layer provides end-to-end delivery (routing) Provides consistent datagram abstraction:

best-effort delivery

no error detection on data

consistent maximum datagram size

consistent global addressing scheme

Link layer (second in OSI) networks provides delivery within the same network

Typically includes its own addressing format (e.g. Ethernet), and maximum frame size (MTU)

17 | 44

Communication Systemsinternet protocol details

IP version 4 is current IPv6 forthcoming

solution to the address space exhaustion of IPv4

Predicted for a while, but limit not reached yet Solutions for preserving numbers in IPv4 (masquerading, private

networks ...) at the moment nobody exactly knows when it will be used other then

in backbone structures

predicted within the next 10 years

3G/UMTS mobile telephone market may push IPv6

defined for a while – we will spend a dedicated lecture on it

18 | 44

Communication Systemsinternet protocol details

Protocol header includes: Version field

Source and Destination addresses

Lengths (header, options, data)

Header checksum

Fragmentation control

TTL, and TOS info

But TOS info often ignored easy changeable along the path (so what for?)

19 | 44

Communication Systemsinternet protocol header details

Version field (4 standard, 5 STII, 6 next gen IP) and IP header length are of 4 byte

IP header normally consists of 20Byte, with options more

Length needed to compute where next header starts

20 | 44

Communication Systemsinternet protocol header details cont.

IP options may sum up to 40Byte Total length field is 16bit, maximum packet length therefore may

not exceed 64kByte Minimum is 20Byte (just the IP header)

MTU of standard physical networks much smaller (e.g. 1,5kByte)

16bit identification field for fragments Set for every packet by original sender of datagram

Sender can not know if fragmentation may occur

Initial message segmentation may not small enough

Copied into each datagram during fragmentation

21 | 44

Communication Systemsinternet protocol header – fragmentation

Content of 16bit identification field is computed by sender Different OS use different computing schemes (tool “nmap” in

practical course)

Might give away information on OS, internal network structure

Masqueraded machines could be identified by their fragmentation IDs

Counter on every machine will have different values (amount of traffic generated, computing scheme ...)

A private network may give more information away than intended

22 | 44

Communication Systemsinternet protocol header – fragmentation

Flags for fragmentation control MF: more fragments (follow)

DF: dont fragment (some protocol implementation like DHCP in Boot-ROMs are not able to reassemble fragmented packets), feature may be used for MTU path discovery (increment packet size until ICMP error message is generated because auf fragmentation need)

Fragmentation offset Offset of this fragment into the original datagram

Zero if no fragmentation used

why offset and not fragment number? - if further fragmentation is needed

23 | 44

Communication Systemsip header – protocol field

Protocol field – the payload with headers removed is passed to a higher layer in the networking stack -> where?

There are different transportation layer protocols for different purposes

1: ICMP – discussed later this lecture 6: TCP – Transmission Control Protocol 17: UDP - User Datagram Protocol 50: ESP – Encapsulating Security Payload 51: AH - Authentication Header All protocol names and corresponding numbers are listed in

(/etc/)protocols file (Linux operating system – see practical course, check on your machine for yourself)

24 | 44

Communication Systemsip header – protocol field

In general: each layer has to provide the information which upper layer should process a given type of packets

Each protocol adds its own header to the packet

25 | 44

Communication Systemsip header – address fields

Source address 32 bit length defines the IP address space

should never be changed through ordinary routing (there are some exceptions like network address translation (NAT))

protocol does not force authentication of source (often enforced by modern routers now)

Destination address 32 bit length defines the IP address space

should never be changed through ordinary routing

changes when source routing used (realised through IP option header)

26 | 44

Communication Systemsinternet protocol – header details

Source routing Special handling for particular datagrams, sometimes don't take

router's "fast path"

Rarely used, but the more common are: Loose Source Routing, Strict Source Routing and Record Route

Timestamp

Must copied on fragmentation

More on routing little bit later on

27 | 44

Communication Systemsip – fragmentation of packets

Gave short introduction already, but ... Adapting datagram size one of the most important tasks of the

Internet working protocol: IP datagrams itself cannot exceed 64kbyte Lower protocol levels report MTU (max. transfer unit)

Linux loopback 16384byte

Ethernet frames offer max. payload of 1500byte

ATM offers 48byte

slow modem-ppp connections 296byte packet length The tool ifconfig or ip (first practical course) reports MTU of each

interface

28 | 44

Communication Systemsip – fragmentation of packets

Fragmentation & Reassembly divide network-layer datagram into multiple link-layer units, all have

to be equal or smaller then link MTU size

further fragmentation may be needed if MTU is decreased along the path again

sometimes it is more clever to set MTU smaller at source to avoid later fragmentation

reconstruct datagram at final station

Each fragment otherwise acts as a complete, routeable datagram Datagrams are identified by the (source, destination,

identification) triple Concept of fragmentation changes with IP v6

29 | 44

Communication Systemsip – fragmentation of packets cont.

If fragmented, identification triple is copied into each resulting packet

Also contains (offset, length, more) triple more - boolean indicates is last fragment

offset - relative to original datagram Relating fragments to original datagram provides:

Tolerance to re-ordering and duplication

Ability to fragment fragments (!)

30 | 44

Communication Systemsip – fragmentation of packets cont.

IP fragments are re-assembled at final destination before datagram is passed up to transport layer

Routers do not reassemble fragmented datagrams Allows for independent routing of fragments

Reduces complexity (need for CPU and memory) in routers Problems with fragmenting:

Loss of 1 or more fragments implies loss of datagram at the IP layer

IP is best effort, provides no retransmission, will time-out if frag(s) appear to be lost

May be interesting for DoS attacks

31 | 44

Communication Systemsip – fragmentation of packets cont.

Avoid fragmentation through computing path MTU Problems if path changes (dynamic routing) and new path has

smaller MTU along its way Adapting size of packets in the source machine according to the

“minimum MTU”: Path MTU Discovery IP v6 uses MTU discovery and assumes standard minimum

MTU If datagram size is smaller then MTU, no fragmentation needed How to do this?

Probe network for largest size that will fit

If possible, have network tell us this size

Operates through ICMP messaging (presented later on)

32 | 44

Communication Systemsip – addressing scheme

We saw that IP packet header reserved two 32 bit fields for source and destination address

For computation for delivery decisions the binary form is used only

Programs and operating systems implementing IP automatically convert the addresses between the two representations

IP addresses are topologically sensitive Interfaces on same network share prefix

Prefix is assigned via ISP/local network administrator

32-bit globally unique

33 | 44

Communication Systemsip – addressing scheme cont.

Address is split into two virtual parts: network and host part See later how division is done

For better reading the binary representation could be split into four octets, which are transferred into the decimal system

34 | 44

Communication Systemsip – addressing scheme cont.

The early IP standard defined five address classes: A, B, C, D and E

An IP address should be selfexplanatory, it should countain information on the networking sub structures History by now

In this view the address consists of a pattern of high order bits, which shows their class, the network and the host component

Machines in the same network share a common prefix (the class definition and network component of IP) and must have unique suffix (the host component of IP)

35 | 44

Communication Systemsip – (historic) address classes

36 | 44

Communication Systemsip – address classes cont.

Class A: (high order bits: 0) Large Organizations, few nets (127), huge number of hosts (16.7

million)

Address range in decimal notation 0.0.0.0 – 127.255.255.255

Class B: (high order bits: 10) Medium sized organizations and firms, e.g. University of Freiburg,

some nets (16,384) and large number of hosts (65,536)

Address range 128.0.0.0 – 191.255.255.255

Class C: (high order bits: 110) Small organizations and firms, relatively large number of nets

(2,097,152) with a small number of hosts per net (256)

Ranging from 192.0.0.0 – 223.255.255.255

37 | 44

Communication Systemsip – address classes cont.

Class D: (high order bits: 1110) Multicast addresses, but service are not very often used

Address range 224.0.0.0 – 239.255.255.255

Class E: (high order bits: 1111) Declared for experimental use only

Address range 240.0.0.0 – 255.255.255.255

Theoretical address space is 4,294,967,296 (seems a lot :-) - but population on earth is higher by now)

But the address space usable for the “Internet” is limited to addresses from 1.X.Y.Z up to 223.X.Y.Z

38 | 44

Communication Systemsip – addressing scheme

But you will loose some more addresses: Special addresses like:

0.0.0.0 defines the default route (explained later, route for the “whole Internet”) or the start address of a host searching for a dynamically provided IP

255.255.255.255 local broadcast address (and destination for hosts seeking an IP via DHCP)

127.0.0.0 loop back network address (you will need only one address within this range and use typically 127.0.0.1). This address is used by every host implementing IP (software using IP for communication is usable without Internet connection)

169.254.0.0 ... 169.254.255.255 is reserved address space for IP ad-hoc/auto configuration without a central server like DHCP

39 | 44

Communication Systemsip – “private” addresses

Addresses reserved for “private” use – many organizations, enterprises, flat-sharing communities need IP communication for their applications without or restricted internet access 10.0.0.0 – 10.255.255.255 (within the class A range)

172.16.0.0 – 172.31.255.255 (16 class B networks)

192.168.0.0 – 192.168.255.255 (65,536 class C networks) University WLAN, private LAN is using 10.X.Y.Z addresses Addresses within these ranges should be discarded on Internet

routers Address classifying helped in the beginning for faster network

decision computation, routers had limited memory and CPU power

40 | 44

Communication Systemsip addressing

For addressing whole subnets or addressing all hosts within a given subnet (possibility depends on the underlying physical network) special IP addresses are introduced

Network number is the smallest IP address in a given (sub)network. it does not address a single machine and may not assigned to a host. It is used with routing tables (explained later in detail)

Broadcast address is the largest possible IP in a network. It should be not assigned to a host, but is the possibility to reach all hosts in a network with just one packet

If we use the example class B address 172.31.5.200, this machine is a member of a network with the network number 172.31.0.0 and a broadcast address 172.31.255.255

41 | 44

Communication Systemsip subnetting

Networks with huge number of hosts could be split into subnets for better administration and considerations on physical topology and global spanning net

The example class B network 172.16 with 65536 host IP numbers in it, allows 256 subnetworks with 256 hosts in it if split on the byte boundary

But: The resulting 256 “class C networks” have the same high order bit like the original class B network

42 | 44

Communication Systemsallowed ip addresses

Not all IP addresses shown are allowed in either packet source or destination address field

0.0.0.0 OK as source, never as destination

255.255.255.255 never as source address, OK as destination

Any network prefix with 255 suffix (might vary for super/subnetting) never as source address, OK as destination

Any network prefix with 0 suffix (network id - might vary for super/subnetting) never as source and destination address

Your router / firewall should filter these combinations

43 | 44

Communication Systemsend of lecture / literature list

Network Layering: Kurose & Ross: Computer Networking (3rd), Section 1.7

Tanenbaum: Computer Networks (4th), Section 1.4

Internet Protocol 4 Kurose & Ross: Computer Networking (3rd), Section 4.4.1

Stevens: TCP/IP Illustrated Vol. 1, Section 3.2

Tanenbaum: Computer Networks (4th), Section 5.5.7, 5.6.1

IP Addressing Kurose & Ross: Computer Networking (3rd): Section 4.4.2

Tanenbaum: Computer Networks (4th): Section 5.6.2

Stevens: TCP/IP Illustrated Vol.1, Section 1.4, Section 3.4

44 | 44

Communication Systemsliterature/exercises

ADSL www.iol.unh.edu/services/testing/dsl/training/ADSL_Tutorial.pdf

Theoretical exercises please hand back the sheet #1 for collection by Mohannad Zalloom

(Hiwi for this course - zalloom@informatik.uni-freiburg.de)

sheet #2 is due to the Tuesday lecture after the pentecost semester break 20th May

check the homepage of this course on the professorships website for the “fast track practical exercise” (on DNS/NSTX)