lecture_2

Lecture 2: Basic routing, ARP, and basic IP

• Literature:– Forouzan, TCP/IP Protocol Suite: Ch 6 - 8

Internetworking

lecture_2

Basic Routing

Delivery, Forwarding, and Routingof IP packets

lecture_2

Connection-oriented vs Connectionless• Connection-Oriented Services

– The network layer establishes a connection between a source and a destination

– Packets are sent along the connection.– The decision about the route is made once at connection

establishment– Routers/switches in connection-oriented networks are stateful

• Connectionless Services– The network layer treats each packet independently– Route lookup for each packet (routing table)– IP is connectionless– IP routers are stateless

lecture_2

Direct vs Indirect delivery• Direct delivery

– The final destination is connected to the same physical network as the sender.

– IP destination address and local interface has same netmask

– Map IP address to physical address: ARP

• Indirect delivery– From router to router, last delivery is direct– Destination address and routing table: Routing

direct delivery

R1

A

R2

R3

B

indirect delivery

indirect delivery

indirect delivery

A B

R

direct delivery

direct delivery

lecture_2

Next-hop Routing• How do you hold information about route from A to all other hosts?

– A R1 R2 R3 B

• Store table of host/network address and nexthop in every node

A

C D E F

BR1 R2 R3

R4N1

N2 N3

N4

N1, -N2, R1N3, R1N4, R1

N1, -N2, R2N3, R2N4, R2

N1, R1N2, R4N3, R4N4, R3

N1, R2N2, R2N3, R2N4, -

N1, R3N2, R3N3, R3N4, -

lecture_2

Routing Table Search - Classful

• Determine class from destination address• Search within class• Routing table often divided into ”buckets”

destination IP addressdestination IP address

Class A bucket

Class B bucket

Class C bucket

lecture_2

Routing Table Search - Classless• Longest prefix first

• Conceptually: divide table in 32 ”buckets” - one for each netmask length and match destination with longest prefixes first

• SW algorithms: tree, binary trees, tries (different data structures)• HW support: TCAMs – Content Addressable Memory

Netid

Netid...

0

1

32

31

Masklen

destination IP addressdestination IP address

lecture_2

Routing Tables• The basic idea with IP addressing (and CIDR) is to aggregate

addresses– more specific networks (with longer prefixes)

less specific networks (with shorter prefixes)

• More aggregation leads to smaller routing tables• The ideal situation is to have domains publishing (exporting) only a

small set of prefixes– Effective address assignment policy

• Some mechanisms lead to increased fragmentation– # of available addresses decreasing distribution of long prefixes (/24)

– Multihoming - sites having several subnetworks – from different providers

• Current routing tables (# of entries) is ~150000 (~60% are /24 prefixes)

lecture_2

Routing Table – Common Fields

• Mask – netmask applied for the entry [255.255.255.0]• Network address – destination network [192.168.15.0]• Next-hop address – next router [130.237.15.1] • Interface – outgoing interface [eth0]• Flags – status/info [U(p), G(ateway), H(ost-specific)...]• Reference count – # of users using this route• Use – # of packets transmitted for this destination

..................................................................................

UseReferencecount

FlagsInterfaceNext-hopAddress

NetworkAddress

Mask

lecture_2

IP Router Model

• A Router can be partitioned into a dataplane and a controlplane– The dataplane is fast and special purpose – handles packet

forwarding in real-time

– The control plane is general purpose– handles routing in the background

IPForwarding

EthernetInterface

FDDIInterface

Router

FIB

RIB

Routing Information

Base

Forwarding Information

Base

IPRoutingControl

Plane

Data Plane

lecture_2

IP Forwarding• A router switches packets between network interfaces• Extracts header information from the incoming datagram

– Destination IP address

• Makes a lookup in the forwarding information base by making a match against networks– Next-Hop IP address,– Outgoing interface,...

• Modifies datagram header• Sends on outgoing interface• But a router performs much more than IPv4 lookup

– Access lists, filtering– Traffic management– Other protocols: Bridging, MPLS, IPv6, ...

lecture_2

ARP

Mapping between logical IP addresses and physical addresses

lecture_2

Logical and Physical Addresses

• A host’s network interface card (NIC) has:– a hardcoded, physical MAC address

• e.g., 48-bit Ethernet address

– a configured, logical IP address

– a configured name

0:0:c0:6f:2d:40 0:0:c0:c2:9b:26 140.252.13.35 140.252.13.34

8:0:20:3:f6:42140.252.13.33

Name:

MAC addr:IP addr:

bsdibsdi sunsun svr4svr4

lecture_2

Communicating with a next-hop

• Problem: bsdi wants to send an IP packet to svr4– No routers between sender and receiver – directly connected host

• Getting the IP address of svr4– Static configuration– DNS: Name Address (Later lectures)

• Getting the MAC address of svr4– Static configuration– Dynamic Address Resolution - ARP

bsdibsdi sunsun svr4svr4

0:0:c0:6f:2d:40 0:0:c0:c2:9b:26 140.252.13.35 140.252.13.34

8:0:20:3:f6:42140.252.13.33

Name:

MAC addr:IP addr:

lecture_2

ARP - Address Resolution Protocol

• Problem: we are to send a packet to an interface on a directly attached network - we know the IP-address of the destination but not the MAC address.

• Idea: Broadcast a request - “On which MAC address can IP-address X be reached?”.– ARP request

• The host/router with the destination replies with its MAC address– ARP reply

• This is the basic functionality of ARP

lecture_2

ARP Examplebsdi intends to send an IP datagram to svr4 (140.252.13.34)

1. Send an ARP request on broadcast to all stations:– who has 140.252.13.34?

2. svr4 identifies it as its own address and sends an ARP reply on unicast back to bsdi– I have 140.252.13.34 and its mac address is 0:0:c0:c2:9b:26

3. bsdi sends the datagram to svr4 using the resolved mac address4. Note that sun and svr4 can update their ARP caches with bsdi!

sun svr4bsdi

1 23

lecture_2

ARP Packet• Two length fields

– Hardware (Ethernet address length: 6)

– Protocol (IP address length: 4)

• Sender Ethernet and IP address

• Target Ethernet and IP address• ARP is encapsulated directly into a data link frame (e.g., Ethernet)

senderEthernet addr

target Ethernet addr

senderIP addr

targetIP addr

1 1 6 64 4

hardware size

protocol size

2

ophwlen

protlen

hwtype

prottype

22

lecture_2

ARP Optimizations• ARP cache

– Resolved addresses are saved in a cache.

– Works because of correlations in use of addresses

– Limits ARP traffic

• Entries in the ARP cache times out

• Network is snooped– Since the sender’s Internet-to-Physical address binding is in every

ARP broadcast; (all) receivers update their caches before processing an ARP packet

lecture_2

ARP Timeouts• If there is no reply to an ARP request

– The machine is down or not responding

– Request was lost, therefore retry (but not too often)

– Eventually give up (When?)

• ARP cache timeouts

– completed entry in 20 minutes (BSD Unix)

– incomplete entry in 3 minutes (BSD Unix)

lecture_2

Indirect/Direct Delivery and ARP• A sends an IP packet to B through router R• Ethernet links to connect A and B to R

IP A IP R IP B

MAC a MAC r1 MAC r2 MAC b

IP Header Src: A, Dst: B Src: A, Dst: B

Ethernet Header Src: a, Dst: r1 Src: r2, Dst: b

Indirect delivery Direct delivery

lecture_2

Proxy ARP (RFC 826)• Proxy ARP - someone

responds to ARP requests on someone else’s behalf

• Allows sub-networks to be hidden

• Example: sun is hidden behind netb: Netb responds on behalf of sun.

gemini

netb

sun

slip

140.252.1.183

140.252.1.129

arp request for 140.252.1.129

arp reply

lecture_2

Gratuitous ARP• Host sends an ARP request of its own address

– Generally done at boot time to inform other machines of its address (possibly a new address) - they get a chance to update their cache entries immediately

– Lets hosts check to see if there is another machine claiming the same address ⇒ “duplicate IP address sent from Ethernet address a:b:c:d:e:f”

• As noted before, hosts have paid the price by servicing the broadcast, so they can cache this information - this is one of the ways the proxy ARP server could know the mapping

• Note that faking that you are another machine can be used to provide failover for servers

lecture_2

RARP: Reverse Address Resolution Protocol (RFC 903)

• How to get your own IP address, when all you know is your link address

• Necessary if you don’t have a disk or other stable storage

• RARP request - broadcast to every host on the network (i.e., EtherDST=0xFFFFFF), TYPE=0x8035

• RARP server: “I know that address!” and sends an RARP reply

• Source host - receives the RARP reply, and now knows its own IP addr

• RARP packet has exactly the same format as ARP packet

• BOOTP/DHCP is a more powerful alternative to RARP

lecture_2

RARP Server• Someone has to know the mappings - quite often this is in

the file “/etc/ethers”• Since this information is generally in a file, RARP servers

are generally implemented as user processes• Unlike ARP responses which are generally part of the

TCP/IP implementation (often part of the kernel)• How does the process get the packets - since they aren’t IP

and won’t come across a socket?– PCAP – Packet Capture (used by Tcpdump/Ethereal)– BPF – Berkeley Packet Filter (older)

• RARP requests are sent as hardware level broadcasts -therefore are not forwarded across routers

lecture_2

IP

Basic functionality and the IP packet header

lecture_2

Issues in IP• Following the end2end argument, only the absolutely

necessary functionality is in IP– Best Effort Service: Unreliable and Connectionless

– Application or Transport layer handles reliability

• How to deliver datagrams over multiple links (hops) in an internetwork?– Addressing– Best-effort delivery service

• Forwarding of packets from one link to another

– Error handling

lecture_2

IPv4 Header – RFC 791• Version

• HLEN – Header Length

• Type of Service

• Total Length– Header + Payload

• Fragmentation – ID, Flags, Offset

• TTL – Time To Live– Limits lifetime

• Protocol– Higher level protocol

• Header checksum

• IP Addresses– Source, Destination

• Options©The McGraw-Hill Companies, Inc., 2000

lecture_2

The Version Field• Version 3 (IEN 21)

– Stems from when TCP was being split into one component handling hop-by-hop communication (IP) and one component handling end-to-end communication (TCP). IEN 21 1 February 1978.

• Version 4 (RFC 791)– IPv4

• Version 5 (RFC 1190)– ST-II - Multimedia streaming protocol

• Version 6 (RFC 2460)– IPv6

lecture_2

The Length Fields• Header Length (4 bits)

– Size of IPv4 header including options.

– Expressed in number of 32-bit words (4-byte words)

– Min is 5 words (=20 bytes)– Max is 15 words (=60 bytes) – limited size limited use

• Total Length (16 bits)– Total length of datagram including header.– If datagram is fragmented: length of fragment.

– Expressed in bytes.• Max: 65535 bytes. (This is IPs length limit)

• Many systems only accept 8K bytes.

lecture_2

The Type of Service Field• Type of Service (ToS): 8 bits• Intended as a field for specifying Quality of Service on a

per-packet basis.• Few applications set the TOS field.

– Unless an added cost/policy check/… associated with usage of a precedence level - it is very likely going to be abused.

• Long history of experimental use– RFC 791 – original– RFC 1122, 1349, 1455 modified the meaning of the ToS field– Current proposal: RFC 2474

• Differentiated Services

– Early Congestion Notification (ECN): RFC 2481, 3168

lecture_2

The ToS Byte – Original proposal

• Original Proposal – RFC 791• Bits 0-2: Precedence

– Defines priority e.g., when packets must be dropped

• Bits 3-5: TOS– Bit 3: 0 = Normal Delay, 1 = Low Delay– Bit 4: 0 = Normal Throughput, 1 = High Throughput

– Bit 5: 0 = Normal Reliability, 1 = High Reliability.

Precedence TOS

Bit 0 Bit 7

lecture_2

DSField – Current Proposal

• Differentiated Services (DiffServ) proposes to use 6 of these bits to provide 64 priority levels - calling it the Differentiated Service (DS) field– RFC 2474– Bits 0-6: Differentiated Services CodePoint (DSCP)

• The DSCP is set when entering an area and determines the QoS handling of the IP datagram in the routers within that area– Scheduling – Shaping– Queue Dropping

• Explicit Congestion Avoidance (ECN)– ECN Capable Transport (ECT)– Congestion Experienced (CE)

DSCP

Bit 0 Bit 7

ECN

lecture_2

Fragmentation – MTU

• If the IP datagram is larger than the MTU of the link layer, itmust be divided into several pieces to fit the MTU – this is called fragmentation

©The McGraw-Hill Companies, Inc., 2000

lecture_2

Fragmentation cont’d• Physical networks maximum frame size

– MTU Maximum Transfer Unit.

• A host or router transmitting datagram larger than MTU of link must divide it into smaller pieces - fragments.

• Both hosts and router may fragment– But only destination host reassemble!– Each fragment routed separately as independent datagram

• In effect, only datagram service (e.g. UDP)– TCP uses 576 byte MTU or path MTU discovery

• 3 fields of the IP header concerns fragmentation

lecture_2

The Fragmentation Fields• Identification: 16 bits

– ID + src IP addr uniquely identifies each datagram sent by a host– The ID is copied to all fragments of a datagram upon fragmentation

• Flags: 3 bits– RF (Reserved Fragment) – for future use (set to 0)– DF (Dont Fragment).

• Set to 1 if datagram should not be fragmented.

• If set and fragmentation needed, datagram will be discarded and an error message will be returned to the sender

– MF (More Fragments)• Set to 1 for all fragments, except the last.

• Fragmentation Offset: 13 bits– 8-byte units: (ip ip_frag << 3)– Shows relative position of a fragment with respect to the whole datagram

lecture_2

Fragmentation Example – Offset

©The McGraw-Hill Companies, Inc., 2000

lecture_2

Fragmentation Example – Detailed

IPv4 hdrid=0, DF=0

UDP hdr Data

IPv4 hdrid=n, DF=0

MF=1, off=0

UDP hdr DataIPv4 hdr

id=n, DF=0MF=0, off=185

Data

8 bytes20 bytes

20 bytes 20 bytes8 bytes

1473 bytes

1472 bytes 1 byte

Offset = 185 185x8 = 1480 bytes

MTU = 1500 bytes

lecture_2

The TTL field• TTL - Time To Live: 8 bits• Limit the lifetime of a datagram - avoid infinite loops• A router receiving a TTL>1 decrements the TTL and

forwards it• A TTL <= 1 shall not be forwarded

– ICMP “time exceeded” is returned to the sender (later slide)

• Recommended value is 64• Should really be called Hop Limit (as in IPv6)

– Historically: Every router holding a datagram for more than 1 second should decrement the TTL by the number of seconds.

lecture_2

The Protocol Field

• Demultiplexing to higher layers

• Assigned by IANA – Internet Assigned

Numbers Authority

• A subset (out of 134) assigned

Reservation ProtocolRSVP46

IPv6 in IPv4IPv641

User DatagramUDP17

Transmission ControlTCP6

IP in IP (encapsulation)IP4

Internet Control MessageICMP1protocolkeyworddecimal

lecture_2

Header Checksum• Ensures integrity of header fields

– Hop-by-hop (not end-to-end)

– The header fields must be correct for proper and safe processing.

– The payload is not covered.

• Other checksums– Link-level CRC. IP assumes a strong L2 checksum/CRC. Hop-by-hop.

– L4 checksums, eg TCP/ICMP/UDP checksums cover payload. End-to-end.

• Internet Checksum Algorithm, RFC 1071– Treat header as sequence of 16-bit integers.

– Add them together

– Take the one’s complement of the result.

lecture_2

Summary• Basic Routing

– Connectionless, next-hop routing

– Routing tables: RIBs and FIBs

– Longest prefix match

• Address resolution– ARP

– RARP

• IP – Internet Protocol– Basic functionality

– Header fields