ece 550d - duke electrical and computer...

ECE 550DFundamentals of Computer Systems and Engineering

Fall 2017

Networking Basics

Prof. John BoardDuke University

Slides are derived from work byProfs. Tyler Bletsch and Andrew Hilton (Duke)

2

Networking

• How do computers communicate?• Two (and only two) computers connected by a direct wire?

• Relatively straight forward: move bits across wire• Internet?

• Many computers, some attached to the same “wire”• All around the world• With other communications going on…• And un-reliable links• And tons of different systems, media, protocols…

• Pretty complicated, so how could we possibly manage it?Abstraction

(oh right, the answer to like…everything)

3

7-layer OSI model

• 7 layer networking stack• (Theoretically) can change out any layer at a time

1 Physical Layer

2 Data-link Layer

3 Network Layer

4 Transport Layer

5 Session Layer

6 Presentation Layer

7 Application Layer

4

Layer 1: Physical Layer

• Defines how our physical media work – wire, fiber, RF• Pin layout• Voltages (or laser pulse description or radio frequency usage)• Timing Requirements

• Examples:• Cat 5 Cable (“Ethernet Cable”) – physical wire• Timing of laser pulses over optical fiber• Wireless radio signal specifications

• Lots of “EE” content, abstracted away from most users and programmers

5

Layer 2: Data-link Layer

• How to move bits across the wires in a meaningful way• Communication between two computers on same physical network• May include some error checking

• Example (but just one of many): Ethernet• Data transmitted in frames• Frame has:

• Pre-amble: used to detect collisions (since >1 machine might try to communicate at once)

• Header: source and destination MAC address• Payload: actual data• CRC check: detect corrupt data

• Carrier Sense, Multiple Access, Collision Detect (CSMACD)• Carrier Sense: listen for if anyone else transmitting• Multiple Access: can wire up many computers to it• Collision Detect: two transmissions at once? Detect and retry random

interval later

6

MAC Address?

• Media Access Control Address• One of two (well, 3 with IPv6) addresses associated with each device

on the modern internet, the other being the IP address(es)• MAC address is unique to each device, more specifically unique to each

network adapter (so wired and wireless are two different MAC addresses) – set forever when device is manufactured (*)

• Used for Layer 2 communication

• ifconfig command

• (*) well, usually…

7

CSMACD

• Ethernet uses CSMACD for multiple systems on a network• Other options, but we won’t go into them• Detection of collisions?

• Pre-amble is fixed pattern• Network card senses medium while transmitting• Mismatch with expected? Collision

• Collision happens?• Exponential backoff• Pick random number of time units• Retry• Fail again? Pick random number from 2x as big a range

• Analogy: crowded dinner party• Try to talk. Someone else talking? Wait. Try again. Fail again? Wait

longer

8

Abstraction: Joys and Limitations

• Joys of abstraction:• Can build an Ethernet card without any info about higher layers• Will work with all of them

• Limitations:• 7-layer model’s abstraction not perfect• Ethernet protocol imposes max limit on cable length

• E.g., layer 2 constrains layer 1• This arises from the need to detect collisions before finishing

sending

9

Midterm Histogram 2017

100 * 90 ** 80 *** 70 ** 6099 89 * 79 ** 69 * <60 ****98 * 88 * 78 ** 68 *97 *** 87 **** 77 ** 67 *96 * 86 * 76 * 66 *95 ** 85 **** 75 * 65 *94 **** 84 *** 74 * 64 *93 * 83 **** 73 6392 * 82 ** 72 *** 6291 *****

***81 * 71 * 61

10

7-layer OSI model

• Reminder where we are so farCat 5 Cable

Ethernet

1 Physical Layer

2 Data-link Layer

3 Network Layer

4 Transport Layer

5 Session Layer


7 Application Layer

11

Our messages so far

• Header says what network layer protocol the pay load is

Preamble Header Payload CRC

12

Layer 3: The Network Layer

• Layer 2 lets computers on same network talk, using just MAC addresses

• Layer 3 lets computers talk across networks• Addressing

• How do we specify what computer to talk to?• Routing

• How do we get from here to there?

• Example: IP protocol• IPv4 and IPv6: pretty similar in most core regards• Best effort delivery• Addressing (IP addresses)• Routing• Analogy: Mailing a letter

13

IP addresses

• IPv4 addresses: 32-bit numbers• Could write as decimal or hex...but that’s not how it’s done• Instead write as four separate bytes (each expressed in decimal),

separated by dots: dotted decimal notation• IPv4 address: 152.3.34.5• As four hex bytes: 98, 03, 22, 05• As a 32-bit hex value: 98032205• 32-bit binary value: 10011000000000110010001000000101

• IPv6: 128-bit addresses• More bits needed to address more separate things on the internet• Been around for a long time, seeing very slow adoption• Addresses represented as eight two-byte groups in hex separated by

colons, for example 3001:0db3:0000:0032:0000:3a2e:0330:7384• We’ll use IPv4 to explain for now

14

IP

• Computer 1 wants to send data to Computer 2• For now, assume it knows IP address (we’ll see DNS later)• Has direct connection to its ISP… but then what?

• The internet is a big place after all..

Computer 166.220.152.32

Computer274.125.130.105

The InternetISP 2

74.125.130.1

ISP 166.220.152.1

15

Let’s zoom in on ISP 1

• ISP has connections to a handful of other places• Generally very high bandwidth connections• Will send your data (packet) to one of these, but which one?

Computer 166.220.152.32

ISP 166.220.152.1

Other Cust A 66.220.152.33

Other Cust B 66.220.152.34

ISP 498.139.183.1

ISP 5

ISP 6129.42.60.1

16

IP Routing

• IP addresses are hierarchical• May not know how to find 74.125.130.105…• But know which way to go to get to 74.______• Move one step closer• Within 74 network, know how to find 74.125• Then 74.125.130• Then find 74.125.130.105

• Analogy:• How do I get to 2200 Mission College Blvd, Santa Clara, CA?

17

IP Routing

• Analogy:• How do I get to 2200 Mission College Blvd, Santa Clara, CA?

• I have no idea, but I can get you to I-40 West• Then you can ask someone else when you get to CA

• Once in CA, you ask someone else:• “I only know its north of here, so take the 5 North and ask

someone else”

• Etc..• This works because our physical addresses are hierarchical:

Country, State, City, Street, Number

18

Routing Basics

• Routing is done with tables• CIDR notation: 40.1.0.0/24

• Match first 24 bits of 40.1.0.0, ignore remaining 8 bits• Find match, entry tells what link to send out on• Example

• 40.0.0.0/8 => Link 0 (match first 8 bits, ignore remaining 24)• 50.1.0.0/16 => Link 1• 50.3.27.0/24 => Link 3 • 50.2.0.0/16 => Link 2• 50.3.42.0/24 => Link 1

19

Routing: More Complex

• Approach one: Static Routing• Enter all routes• Let system run• Hope nothing goes down• Works fine for small networks

• Reality:• Network links/systems go down• Often multiple paths to same place

• Changing traffic patterns = changing fastest route

20

Distance Vector Protocols

• Routers• Know distances to immediate neighbors• Compute distance vector

• How far to any destination from all known info• Transmit distance vectors to neighbors

• Discover better (shorter) route? Update table• Now know more info, so repeat process

21

Distance Vector Routing

A

C

Dz

Dest Cost RtA ∞ —B ∞ —C ∞ —D ∞ —

v

Dest Cost RtA 2 AB ∞ —C 2 CD ∞ —

x

Dest Cost RtA ∞ —B 2 BC ∞ —D ∞ —

w


1

B

t

Dest Cost RtA ∞ —B 1 BC ∞ —D 3 D

2

3

3

2

4

3

2

1

2

2

22


A

C

Dz


v


x


w


1

B

t


2

3

3

2

4

3

2

1

2

2

2+1 via v2+3 via x2+1 via v

23


A

C

Dz


v


x


w


1

B

t


2

3

3

2

4

3

2

1

2

2

no improvements

24


A

C

Dz

Dest Cost RtA 4 vB 3 tC 4 vD 5 t

v


x

Dest Cost RtA ∞ —B 2 BC ∞ —D 6 t

w

Dest Cost RtA 3 vB 5 xC 3 vD ∞ —

1

B

t


2

3

3

2

4

3

2

1

2

2

25


A

C

Dz

Dest Cost RtA 4 vB 3 tC 4 vwD 5 t

v

Dest Cost RtA 2 AB 5 zC 2 CD 7 z

x

Dest Cost RtA 6 wB 2 BC 6 wD 6 t

w

Dest Cost RtA 3 vB 5 xC 3 vD 9 x

1

B

t

Dest Cost RtA 6 zB 1 BC 6 zD 3 D

2

3

3

2

4

3

2

1

2

2

26


A

C

Dz


v


x


w

Dest Cost RtA 3 vB 5 xC 3 vD 8 v

1

B

t


2

3

3

2

4

3

2

1

2

2

27


A

C

Dz


v


x


w


1

B

t


2

3

3

2

4

3

2

1

2

2

Now everything is stable…but what if…

28


A

C

Dz


v


x


w


1

B

t


2

3

3

2

4

3

2

1

2

2

Router z fails?

29


A

C

Dz


v

Dest Cost RtA 2 AB 6 wC 2 CD 9 w

x


w


1

B

t

Dest Cost RtA 9 xB 1 BC 9 xD 3 D

2

3

3

2

4

3

2

1

2

2

Temporary problem:v <->w routing loop!

30


A

C

Dz


v


x


w


1

B

t


2

3

3

2

4

3

2

1

2

2

Fixed it!(v has stale cost still)

31


A

C

Dz


v


x


w


1

B

t


2

3

3

2

4

3

2

1

2

2

32


A

C

Dz


v


x


w


1

B

t


2

3

3

2

4

3

2

1

2

2

Stable again.If z comes back, we’ll rediscover better routes

33


A

C

Dz


v


x


w


1

B

t


2

3

3

2

4

3

2

1

2

2

But what if v fails?

34


A

C

Dz


v


x


w


1

B

t


2

3

3

2

4

3

2

1

2

2

v is down…but x advertises routes to A and C (cost 6)

35


A

C

Dz


v


x


w

Dest Cost RtA 9 xB 5 xC 9 xD 9 x

1

B

t


2

3

3

2

4

3

2

1

2

2

update w’s tableNow x has new info

36


A

C

Dz


v


x

Dest Cost RtA 9 wB 2 BC 9 tD 6 t

w

Dest Cost RtA 9 xB 5 xC 9 xD 9 x

1

B

t


2

3

3

2

4

3

2

1

2

2

Update x’s table..w and t will update to 12(routing through x)

37

Count-to-infinity

• Algorithm will slowly “count to infinity”• Actually: count to max value it holds• Then throw away the route, concluding there is no path there

• Packets sent in meanwhile? • IP: Time To Live (TTL)• Starts at fixed value (e.g., 255)• Decremented every time the packet is forwarded• Packet dropped when TTL == 0

• Traceroute: uses TTL fields to probe to different distances• Uses ICMP protocol to get response on TTL

• Note: many fancier routing schemes, we aren’t covering

38


x

Dest Cost RtA 9 wB 2 BC 9 tD 6 t

w

Dest Cost RtA 9 x—B 5 xC 9 xD 9 x

3

Obvious optimization:- If x gets a route from w, it should not advertise that route back to w- Called “split-horizon”- Helps stabilize faster, but does not solve the problem

(may still count to infinity)

39

Link State Protocols

• Another option: link state protocols• Send info about direct connections to all routers• All routers build global pictures of network• Run graph algorithms to find shortest paths

• E.g., Dijkstra’s shortest path algorithm• Global information is nice, but…

• Complex for very large systems• How many routers on the internet?• Do they all exchange all their info and run Dijkstra’s?

• Of course not..• So… what do we do?

39

Use Abstraction… (and hierarchy)

40

• Divide internet up into Autonomous Systems (ASes)• Each AS can advertise routes to other ASes• Routing internal to AS is hidden from outside world

• Can be Link State, Distance Vector, other…• We won’t go into too many details

40

Border Gateway Protocol

AS 1

AS 4

AS 2AS 3

41

The IPv6 problem

• IPv4 developed in late 1970’s, deployed in early 1980’s• Computers were big and expensive; there was one network computer

for the entire School of Engineering then (I ran it!).• PCs had just been invented but it was silly to think they would ever be

on the internet, which was for serious work• So 2^32 addresses was an approximation of infinity• Initial tenants (most in US) got huge allocations – MIT alone got

multiple /8 networks.• (Global allocations, MIT sale)

42

7-layer OSI model

• IP: hierarchical addresses + best effort deliveryCat 5 Cable

Ethernet

1 Physical Layer

2 Data-link Layer

3 Network Layer

4 Transport Layer

5 Session Layer


7 Application Layer

IP

43

Our messages so far

• IP Header has• Src IP• Dest IP• Payload type (what protocol)• Other info

• Side note: IP does fragmentation to fit within frame size• We aren’t covering that

Preamble Header Header CRCPayload

44

Layer 4: Transport Layer

• Reliability (if used)• Acknowledgements of data receipt• Retries of failed data for individual packets – does not guarantee

delivered in order for instance!• Flow Control

• Restrict rate of data sending• Multiplexing/De-multiplexing data

• E.g., Ports: identify which program some data is for• Keep data streams separate

45

Layer 5: Session Layer

• Concept of “a connection”• Establish/terminate• (OSI includes a variety of obscure features not often used)

• TCP: combines these two layers together• Sets up/terminates sessions• Has sequence numbering for packets• Acknowledges (ACKs) packets that are received• Establishes flow control (responds to congestion by throttling sending)

46

TCP

• We’ll draw diagram with computers on each side• Time goes down• Three messages above (TCP’s “3 way handshake”)

9987: SYN, ACK(1045)

1045: SYN

1046: ACK(9987)

47

TCP

• To open a new connection:• 1 computer sends SYN (“Hi, lets talk”)

• All messages have sequence numbers including SYN• First sequence number of a new connection is random• TCP sequence numbers by byte

9987: SYN, ACK(1045)

1045: SYN

1046: ACK(9987)

48

TCP

• Other computer• ACKs (Acknowledges) the message (says what sequence # it ACKs)• Also sends SYN “Hey sure, lets talk”

9987: SYN, ACK(1045)

1045: SYN

1046: ACK(9987)

49

TCP

• First computer then ACKs this SYN• And probably sends data along with the ACK

• TCP control info (SYN, ACK, FIN): bits in TCP header• Packets can have multiple control bits on + carry data

9987: SYN, ACK(1045)

1045: SYN

1046: ACK(9987)

50

TCP: Normal operation

• Data going right in blue• ACKS coming left in green

• note: ACK #ed by expected next data• Sliding window (flow control)

• Limit amount of un-ACKed data at a time

200025003000

35004000

4500

ACK(2500)ACK(3000)ACK(3500)

51

TCP: Re-ordered Data

• Data may get re-ordered in network• One packet takes one route, another takes another

• TCP: no problem• Sender re-orders data properly• Sends ACK for as much data as it has

5000

ACK(6000)

5500

52

TCP: Lost data

• Data may also get lost in the network• E.g., router is backlogged, can’t handle it has to drop from queue

• TCP will re-send un-ACKed data after a timeout

6000

ACK(6500)

6000

53

TCP: Duplicate Data

• Receiver may get duplicate data• 7000-7499 gets lost• But 7500-7599 arrives• Then sender re-sends both: no ACK for either (why not?)• No problem: receiver drops duplicate (can tell: sequence #s)

7000

ACK(8000)

7500

7000

7500

54

TCP: Closing connection

• Connection closed with FIN message• Receiver ACKs

• Other side may close (with FIN) [typical]• Or remain open: can still send data• Side that closed cannot send, but should receive/ACK• FIN/ACK may be one message

9000 FIN

ACK(9016)

101090 FIN

ACK(101106)

55

TCP: Closing connection

• What if ACK for FIN gets lost?• FIN gets retried… but other side expects connection is closed?

• TCP has a state to handle this• Connection expected to be closed, but resources/state still held• Times out if no activity (assumes ACK got through if no retry)

9000 FIN

ACK(9016)

101090 FINACK(101106)

101090 FINACK(101106)

56

Flow control: Sliding Window

• Problem:• Congestion -> dropped packets• Dropped packets -> Retries• Retries = duplicates of data -> More congestion

• Vicious cycle…

• TCP implements flow control with a sliding window• Limitation of amount of un-ACKed data out at a time• Retry required? Shrink window

• Assumes congestion, tries to avoid it• No retries in a while? Grow window back

• Maybe it cleared up?

57

7-layer OSI model

• TCP: It’s the coolest thing since memory got sliced into pages!Cat 5 Cable

Ethernet

1 Physical Layer

2 Data-link Layer

3 Network Layer

4 Transport Layer

5 Session Layer


7 Application Layer

IP

TCP

(TCP)

58

Our messages so far

• TCP header has• Source/Dest port• Sequence numbers• Control bits (SYN/ACK/FIN)• Check sum over data• Other stuff

Preamble Header Header CRCHeader Payload

59

Layer 6: Presentation Layer

• Responsible for data formats• Examples

• Character encoding schemes• Serialization of objects

• We’re not really going to talk about it much

60

Layer 7: Application Layer

• Protocol specific to how applications want to communicate• Examples:

• HTTP(S)• (S)FTP• SSH• SMTP• IMAP• …

• Again, not going into this much…

61

7-layer OSI model

• Flexibility Example: Wired vs Wireless• Change out two layers, rest stay the same

Cat 5 Cable

Ethernet

IP

TCP

(TCP)

-

HTTP

Wireless Radio

802.11

IP

TCP

(TCP)

-

HTTP

62

7-layer OSI model

• Flexibility Example: A different application on top of bothCat 5 Cable

Ethernet

IP

TCP

(TCP)

-

SMTP

Wireless Radio

802.11

IP

TCP

(TCP)

-

SMTP

63

How to find out IP addresses?

• I don’t know about you, but whenever I want to visit Google, I just type “172.217.2.206” into my browser

• Not really• People don’t memorize IP addresses – need convenient way to

translate human-readable names to IP addresses• This is the domain name system (DNS)• Hierarchical servers with hierarchical database:

Root DNS Servers

com DNS servers org DNS servers edu DNS servers

poly.eduDNS servers

umass.eduDNS serversyahoo.com

DNS serversamazon.comDNS servers

pbs.orgDNS servers

Figure from DNS – Domain Name System, CS234, UC Irvine.

64

requesting hostcis.poly.edu

gaia.cs.umass.edu

root DNS server

local DNS serverdns.poly.edu

1

2

45

6

authoritative DNS serverdns.cs.umass.edu

7

8

TLD DNS server

3

DNS name resolution example

Recursive query:• Resolves any query by

consulting servers that are authoritative for the query

• To do so, servers may recursively query other servers higher up in the hierarchy

Adapted from DNS – Domain Name System, CS234, UC Irvine.

65

Additional DNS facts

• DNS contains multiple kinds of records:• Name -> IP translations (“A” records)• IP -> Name translations (“PTR” records)• Name -> Name translations (“CNAME” records)• more

• DNS clients and servers can cache results• The local Duke DNS server definitely has “google.com” cached

• Second-level domains (“duke.edu”, “google.com”, etc.) are registered, usually by paying a fee, with a registrar (e.g. “easydns.com”) who interacts with the top-level domain authority (Verisign for .com, Educause for .edu, etc.)

• Sub-domains can be created administratively by the owner of a domain (e.g. “whatever.duke.edu” can be made by the “duke.edu” admin).

66

Wireshark demo

67

Network programming

• Coding networking code in…• Java: Look in java.net, start with Socket• C:

• socket()• connect()• accept()• bind() • listen()

68

Example C sockets clientDoes a simple HTTP request of google.com

// adapted from simple-client.c by Sean Walton and Macmillan Publishers// adapted by Tyler Bletsch for Duke University

// This program will request http://google.com/ and print the first 1024 bytes of the response// Note: for simplicity, the google IP address is hard coded

#include <stdlib.h>#include <stdio.h>#include <unistd.h> // needed for close#include <string.h> // needed for bzero#include <sys/socket.h> // needed for socket calls#include <resolv.h> // needed for socket type#include <arpa/inet.h> // needed for inet_aton

#define PORT 80#define SERVER_ADDR "172.217.1.14" // ^ this is google.com#define MAXBUF 1024

char request[] = "GET / HTTP/1.0\r\n\r\n";int request_len = sizeof(request)-1;

#define die(s) { perror(s); exit(1); }

int main() {int sockfd;struct sockaddr_in dest;char buffer[MAXBUF];

/*---Open socket for streaming---*/if ( (sockfd = socket(AF_INET, SOCK_STREAM, 0)) < 0 )

die("socket");

/*---Initialize server address/port struct---*/bzero(&dest, sizeof(dest));dest.sin_family = AF_INET;dest.sin_port = htons(PORT);if ( inet_aton(SERVER_ADDR, &dest.sin_addr) == 0 )

die(SERVER_ADDR);

/*---Connect to server---*/if (connect(sockfd, (struct sockaddr*)&dest, sizeof(dest)) != 0)

die("connect");

/*---Send request---*/send(sockfd, request, request_len, 0);

/*---Get and print response---*/bzero(buffer, MAXBUF);recv(sockfd, buffer, sizeof(buffer), 0);printf("%s", buffer);

/*---Clean up---*/close(sockfd);return 0;

}

69

Test run

70

Summary

• Networking Overview• 7-layer model• Emphasis on IP (Layer 3) and TCP (Layers 4 and 5)

• Not comprehensive, but…• You are now at least conversant enough to discuss the OSI stack at

parties

ece 550d - duke electrical and computer...

Documents