The Internetworking Problem

► Two nodes communicating across a “network of networks”… How to transport packets through this heterogeneous mass ? Internetworking A B

Cloud

Cloud

Cloud

A Translation-based Solution The Internetworking Problem…

► Before : different packet-switching networks (e.g., ARPANET, ARPA ) … only nodes on ALG the same network could communicate ALG

ALG ALG

► application-layer gateways ™ difficult to deploy new internet-wide applications ™ hard to diagnose and remedy end-to-end problems ™ stateful gateways inhibited dynamic around failures ► no global addressability ™ ad-hoc, application-specific solutions

1 Native Solution: Directly on IP as an Intermediary Network ► Key ideas:

(FTP – File Transfer Protocol, NFS – Network File Transfer, HTTP – World Wide Web protocol) ™ Overlay: better than any↔any translation. Fewer, simpler mappings. Application Telnet FTP NFS HTTP ™ Network-layer: efficient implementation, global addressing

Coaxial Fiber Packet Transmission Application Telnet FTP NFS HTTP Media cable optic radio Intermediate Layer (IP) ► No network level overlay: each new

application has to be re-implemented for Transmission Coaxial Fiber Packet every network technology! Media cable optic radio

Internet Protocol (IP) Overview

► IP runs over everything ► Networks linked together by Internetwork programs -- each supports them ► Example Internetwork shown next: ™ Consists of two networks – Ethernet and Token ring ™ Host connected to Ethernet can talk to one connected to the Token

2 IP Overview… IP Host Processing

► The two hosts -- are end systems ► Host processing for TCP transfer is shown ► -- intermediate system next: ™ data is sent down the ► Application layer and transport layer (e.g., TCP, UDP) Æ End-to-end protocols ™ each layer adds information to the data -- prepending headers and trailers ► Network layer Æ Hop-by-hop protocols used on the two end systems and every intermediate system

IP Host Processing … IP Host Processing …

► Nearly identical scheme true for other protocols ™ TCP, UDP, ICMP, and IGMP send data to IP ™ IP adds protocol field in the IP packet to indicate what type of data is in the data segment -- ICMP = 1, IGMP =2, TCP=6, UDP = 17 ► Similarly, many different applications can be using TCP or UDP at any one time

3 IP Host Processing… IP Host Processing…

► Network interface sends/receives frames on behalf of IP, ARP, RARP -- a 16-bit frame type field in the Ethernet header is used to indicate the protocol ► Headers are removed as packet goes up ► Each protocol box uses the protocol field to demultiplex the packet among the upper layer receivers

IP Host Processing… : Details

► IP header not only carries Internet source and destination addresses -- specify a number of parameters that are essential for routers

4 IP Details… IP Details…

► Header has fixed fields present in every packets ► Total length: number of bytes contained and several options in the packet -- including the IP header ► Header fields are aligned at 32-bit boundary ► TTL (time-to-live): sets the upper limit on ► Version: 4 = IPv4 the number of routers through which a ► IHL: Internet header length -- in 32-bit words -- datagram can pass the length is 5 (when no options) and varies to ™ each router decrements this by 1 as the 15 -- 40 bytes are allowed for IP options packet passes ► Type of service: defines packet’s precedence ™ when it reaches 0, the packet is thrown away and desired type of routing and an ICMP message is sent to the source

Internet Protocol Internet Protocol

► Identification, ► Routing protocols normally use the “best” route - Flags, - there are several definitions of “best” -- Fragment cheapest, fastest, most reliable offset: used by fragmentation ► Type of service specifies the requirement of the and reassembly application to the routing protocol ► Protocol: used at the destination for demultiplexing the packet

5 Internet Protocol Internet Protocol

► Precedence indicator does not affect Fragmentation and Reassembly: routing but queuing ► Internetworking programs are expected to relay ™ several packets waiting for transmission on packets between heterogeneous networks the same channel ► Network technology implies a maximum packet ™ highest precedence should in theory be size transmitted first ► Incoming packet fragmented: ► The “differentiated services” effort is refining this traditional interpretation of type-of-service

Internet Protocol Internet Protocol

► Incoming fragment fragmented: ► Host can’t reuse an identifier if there is a risk of fragmentation and new fragments mixing with old fragments

► Wait for the expiration of the fragments -- TTL

► With a packet size of 4k, this translates to about 17Mbps transfer rate -- clearly not adequate ► Identification + source address uniquely identifies the packet fragment for the destination ► Correct way is to discovery the path MTU and use it as the max packet size -- no fragmentation ► Receiver assembles all fragments with same ID according to the “offset”

6 Internet Protocol Internet Protocol

Path MTU discovery: IP Options: ► sets the DF (don’t fragment bit) in the IP header to discover if any router on the current path ► IP options field is used to carry specific needs to fragment functions ► ICMP error message is returned by a router ™ request specific routing for some packets, asked to forward an IP with DF set when MTU is e.g., loose source routing, strict source less than datagram size routing ► this error message is used to decrease the ™ in source routing, the send specifies the route datagram size until no error -- TCP path MTU by specifying intermediate routers discovery ► options are rarely used now. Why?

Internet Protocol Internet Protocol

Options and header processing: ► To speed up router processing ► IP options are used rarely because of the ™ optimize the most commonly used case processing costs ™ without the IP options -- the header has five ► naive implementation of IP routing will perform 32-bit words -- makes verification faster ™ verify version field, checksum, compatibility checks, ™ frequently used routes can be cached -- to and parse any IP options achieve Gigabit-per-second routing! ™ look for next hop for destination address considering type of service, interface and so on. ► Packet with options create problems -- ™ takes hundreds of instructions they are processed with low priority than “normal” packets

7 Internet Protocol Internet Protocol

► Because of the performance penalty for ► Cost at router C will be comparable, i.e., options -- alternatives are used for source processing options and unwrapping routing ► Cost at intermediate routers A to C, and C ► One technique is called “encapsulation” to B are less with the encapsulation -- ► Instead of specifying a loose source processed using the optimized methods -- routing “from A to B through C” -- no option processing encapsulate a packet “from A to B” in another packet “from A to C”

Internet Control Message ICMP Details… Protocol

► IP is straightforward and simple –no ► Most ICMP packets are diagnostic info. feedback for diagnosing error conditions sent back when a router destroys a packet ► Internet Control Message Protocol (ICMP) -- e.g., destination unreachable, TTL does this feedback expired ™ layered on top of IP -- protocol type 1 ► ICMP also defines a echo function used ™ all routers and hosts are expected to “speak” for testing connectivity this protocol ► ICMP does not make IP datagram service reliable

8 ICMP Details… ICMP Details…

► Just provides feedback about network ► All ICMP messages problems start with a common 32-bit ICMP header ► ICMP is carried on IP datagrams -- these packets themselves could be victim of errors ► No ICMP error is triggered by an ICMP message

ICMP Details… ICMP Details…

► Reporting “operational” problems such as ► Destination unreachable messages are time exceeded, destination unreachable, sent when a router cannot forward a source quench is the most common use packet ► These packets have the same format -- includes the entire header of and 8 bytes of the triggering packet

► Time exceeded message sent when a packet is destroyed because TTL expired

9 ICMP Details… ICMP Ping

► Source quench messages are sent by a ► When a router/host receives an ICMP message router that detects congestion -- source is of type echo, it responds by an “echo reply” supposed to reduce sending rate when it receives this message ► Parameter problem message is sent by a router that find an error in the encoding of ► Reply is derived from request by swapping the the IP header IP header’s source and destination address -- replacing ECHO by ECHO-REPLY and computing new checksums

ICMP Traceroute ICMP Router Discovery

► Traceroute tries to discover intermediate routers ► To send a packet – a host needs nexthop ► Send packet with TTL = 1; first router ™ Test whether packet destination in current decrements TTL to 0, destroy the packet and subnet send back a “TTL expired” ICMP message ™ If not, forward packet to a router so that ► Source address of the ICMP identifies the first packet can reach the destination router ► With hen there are multiple routers ► Next message is sent with TTL+1 for second connected to the local network -- host router should select the one nearest to the ► Packet sent for an unused UDP port an ICMP destination port unreachable -- message is sent back

10 ICMP Router Discovery … ICMP Router Discovery …

► How to discovery local routers? ► Router advertisements contain a list of ™ Read from a config file – static solution routers with a preference notation ™ Dynamic solution – zero admin overhead – ► Hosts select router with highest preference discovery procedure ► Router discovery using special ICMP messages ► Routers send “router advertisements” at regular intervals ► Hosts trigger this by sending “router solicitations”

ICMP Redirection ICMP Redirection…

► Default route is a nice way to establish ► Default gateway X for A: packets go to connectivity -- may not be very efficient Internet through X ► Packets sent by A to B first go to X and then to Y -- packets cross LAN twice ► “Redirection” allows routers to inform route updates to hosts to correct this problem ► The first packet to B will be sent to X -- it triggers a “redirect” ICMP message from X to A

11