How the Internet Works · What is the Internet Made of? l Computers l Servers l Clients l Phones l “Things” l Routers—specialized computers that forward “packets” l Packets

HowtheInternetWorksStevenM.Bellovin

DepartmentofComputerScience,ColumbiaUniversityhAps://www.cs.columbia.edu/~smb

1

DisclaimerAllofthestatements,opinions,facts,myths,errors,etc.,inthistalkaremineandminealone,anddonotrepresenttheopinionsofColumbiaUniversityorofanyagencyoftheUSgovernment.

2

WhatistheInternetMadeof?l  Computers

l  Serversl  Clientsl  Phonesl  “Things”

l  Routers—specializedcomputersthatforward“packets”l  Packetsarefragmentsofmessages

l  Links—WiFi,Ethernet,fiber,etc.TheInternetwasdesignedtorunoveranything

3

Fibersl  Eachcablehasmanypairsofstrands

l  Eachstrandcarriesmanywavelengths(aka“colors”or“lambdas”)l  Anewtrans-Pacificfiberhassixpairsofstrandsl  Eachstrandcarries100wavelengthsl  Eachwavelengthhasabandwidthof100Gbpsl  Totalcapacity:60terabits/second

l  Eachwavelengthcancarrymanydifferentcircuits

l  EachInternetcircuitcarriespacketsformanydifferentconversa_ons

4

WiFil  Usedinpublicspacesandprivateresidences

l  Someuseinbusiness,butwiredEthernetismorecommonfordesktops

l  Range:about100meters

l  Security:WEPisobsoleteandinsecure;WPA2isquitegood—andinpublic,allbetsareoff.

5

ALookatCommonApplica_onsl  Webbrowsing

l  Email

l  TheCloud

l  Cau1on:allofthisissimplified—andarguablyoversimplified

6

HowtheWebAppearstoUsers

7

Internet

WebBrowser

WebServer

TheInternetHasStructure:Mul_pleISPs

8

ISPA

ISPB

LON

NYC

Rou_ngBetweenISPs

9

9

Verizon

Sprint

IIJ

Big ISPs ‘Peering’

GoJ

Amazon Customers buy ‘Transit’

Sakura

EachISPHasStructure:ManyRouters

10

Hos_ngServices

11

Internet

WebBrowser

Hos_ngCompany

CompanyA

CompanyB

CompanyC

ContentDistribu_onNetwork

12

CDNACDNB

CDNC CDNDWebServer


13

CDNACDNB

CDNC

CDNDWebServer


14

CDNACDNB

CDNC

CDNDWebServer


15

CDNACDNB

CDNC CDNDWebServer

CDNExample:www.supremecourtus.gov

NewYork 24.143.200.48

Ashburn,Va 23.15.9.144

Atlanta 208.44.23.57

SanFrancisco 216.156.149.106

Boston 207.86.164.89

16

www.supremecourt.govisanaliasfora1042.b.akamai.net;AkamaiisaprominentCDNoperator

WhichistheBrowser;WhichistheServer?

17

Internet

WebBrowser

WebServer

Architecturally,They’retheSame—WhatMaAersistheSolwareTheyRun

18

Internet

WebServer

WebBrowser

“SmartHosts,DumbNetwork”l  Thephonenetworkwasbuiltfordumbphones–nothingelsewastechnicallyoreconomicallyfeasible.

l  Allintelligenceisinthenetwork:conferencecalls,callforwarding,evenmanyvoicemenus

l  Internetroutersareverydumb;allintelligenceisinendsystemsl  Consequence:serviceprovidersarenotnecessarilythesameasnetworkprovidersl  Aperson’smailprovidermaybeinanothercountry

19

ThePhoneNetwork:AFewLargeSwitches,ServingPhones

20

TheInternet:ManyRouters,VeryManyTypesofDevices

21

CircuitSwitchingversusPacketSwitchingl  Circuits:tradi_onaltelephonymodel

l  Paththroughthenetworkselectedat“callsetup_me”l  Verysmallnumberofcallsetups;process

canbeheavyweight

l  Each“phoneswitch”needstoknowthedes1na1onofthecall,notthesource;returntraffictakesthereversepath

l  Packets:Internetmodel

l  Every“packet”–afragmentofamessage–isroutedindependentlyl  Nocallsetupl  Rou_ngmustbevery,veryfast;it’sdone

foreachpacket

l  Robustness:ifa“router”fails,packetscantakeadifferentpath

l  Everypacketmusthaveasourceanddes_na_onaddress,toenablereplies

l  Replytrafficmaytakeaverydifferentpath

22

IPAddressesl  Ausertypesanamesuchaswww.dni.gov.

l  TheDomainNameSystem(DNS)translatesthattoanInternetProtocol(IP)Addresssuchas23.213.38.42l  IPaddressesarefourbyteslong;eachofthosenumbersisintherange0-255l  www.dni.govactuallyusesaCDN,soeveryqueriergetsadifferentanswer

l  IPaddressesarewhatappearinpackets

l  Routerstalktoeachother(viaRou1ngProtocols)tolearnwhereeachIPaddressis

23

IPAddressing

l  Roughly4billionpossibleIPaddressestodayl  IPv6,anewerversionofIPbeingdeployednow,hasmanymoreaddresses

l  IPaddressesarehandedoutinblockstobigISPs.BigISPsgivepiecesoftheiralloca_onstosmallerISPsortoendcustomers

l  Unlessyou’reaverylargeenterprise,theonlywaytogetIPaddressesisfromyourISP–andifyouswitchISPs,youhavetorenumberyourcomputers

l  Thereisnoanalogto“localnumberportability”ontheInternet–andcan’tbe;there’sno_metodothatmanylookups

24

AddressSpaceAssignmentl  IPaddressesarehandedoutbyRegionalInternetRegistries(RIRs),suchasARIN

l  TheygettheiraddressesfromICANN,aninterna_onalnon-profitwhichgetsitsauthorityfromtheU.S.DepartmentofCommerce–controversialabroad

l  Addressesareallocatedbasedondemonstratedshort-termneedandevidenceofefficientuseofpreviously-allocatedaddresses

l  Addressesmaynotbesold,evenaspartofabankruptcy,merger,oracquisi_on,exceptwithARIN’sapprovalandinaccordancewithARIN’spoliciesl  Thisasser_onofauthorityhasneverbeencontestedincourt—andsomehavebeentransferredby

orderofabankruptcycourtl  SomeISPshave(veryvaluable)pre-ARINaddresses,called“legacyspace”.Legacyaddressholders

don’thavetorenumberwhenswitchingISPs(amongotheradvantages)

25

PortNumbersl  Whenonecomputercontactsanother,isittryingtotalktoaWebserverortryingtosendmail?l  Rememberthatarchitecturally,allmachinesontheInternetarealikel  It’sperfectlylegaltorunaWebserverandamailserveronasinglecomputer

l  PacketscontainnotjustanIPaddressbutaportnumberl  Port25isthemailserver,port80istheWebserver,443isencryptedWeb,etc.

l  IfanIPaddressislikeastreetaddress,aportnumberistheroomnumberinthebuildingl  Room25isthemailroom,room80islibrary,etc.

26

TheNetworkStackl  TheInternetusesalayered

architecture

l  Applica_ons—email,web,etc.—arewhatwecareabout

l  TCP(whichhasportnumbers)transportsthedata;itisend-to-end

l  IP(thenetworklayer)isprocessedbyeveryrouteralongthepath

l  ThelinklayeristhingslikeWiFi,Ethernet,etc.

27

Email

28

SendingEmail

29

ISP

ISP ISP

ISP

OutboundMailServer

InboundMailServer

AccessLinks

SendingMyselfEmail—AnSMTPTranscript

30

220machshav.comESMTPExim4.82Tue,11Mar201419:43:03+0000HELOeloi.cs.columbia.edu250machshav.comHelloeloi.cs.columbia.edu[2001:18d8:ffff:16:12dd:b1ff:feef:8868]MAILFROM:<[email protected]>250OKRCPTTO:<[email protected]>250AcceptedDATA354Entermessage,endingwith"."onalinebyitselfFrom:BarackObama<[email protected]>To:<[email protected]>Subject:TestThisisatest.250OKid=1WNSaS-0001z5-1dQUIT221machshav.comclosingconnec_on

Message

Conversa_onWithAThirdParty

31


Message

WhattheRecipientSees

32


Message

ALeAerfromEleanorRoosevelttoLorenaHickock(March1933)

33

Itbegins“Hickmydearest”.(excerptfromAmazon.com)

ThingstoNotel  TheSMTPenvelope—that’sthetechnicalterm!—canhavedifferentinforma_onthanthe

messageheaders

l  Unlikethephonenetwork,anyonecanruntheirownmailserversl  Ipersonallyruntwo,onepersonalandoneprofessionall  Thiscomplicatesthirdpartydoctrineanalysis

l  TherealityofemailisfarmorecomplexthanI’veoutlinedherel  Example:manypeoplereadtheiremailviaaWebbrowser—andtheNSAhasstatedthatevenfor

them,pickingoutjusttheFrom/Toinforma_onfromaWebmailsessionisverydifficult

l  Ihaven’tevenbeguntoaddressserver-residentemail,virusscanning,spamfiltering,andthelike,letalonealloftheothermetadatathat’spresent

34

Encryp_onontheInternet

35

AnythingCanbeEncryptedl  Links—thoughmostlyusedonWiFi

l  VirtualPrivateNetworks(VPNs)

l  Simpleconnec_ons(Web,email,etc.),generallyviaTransportLayerSecurity(TLS)

l  Data,especiallythebodyofemailmessages

36

VPNsl  Usedbycorporateemployeesfortelecommu_ngorwhiletraveling

l  Alsousedtoconnectmul_plecorporateloca_ons

l  Some_mesusedtospoofloca_onl  Covertracksl  Foolgeographicrestric_onsoncontent,e.g.,streamingmoviesandmusic

l  ArecentlypublishedacademicpaperconcludedthattheNSAcouldcryptanalyzealotofVPNsessions

37

TLSl  UsedforallsecureWebtraffic

l  Widely(andincreasingly)usedwhensendingandretrievingemaill  But—TLSdoesnotprotectemail“atrest”,i.e.,whileondiskonthevariousservers

l  Usedformanyotherpoint-to-pointconnec_ons,e.g.,Dropbox

l  OlderversionsofTLShavecryptographicweaknesses;theseare(believedtobe)fixedinthenewestversions

l  Themostcommonimplementa_onsofTLShavealonghistoryofserioussecurityflaws

38

EmailEncryp_onl  Twodifferentstandards,S/MIMEandPGP

l  S/MIMEiswidelysupported—butrarelyusedl  PGPrequireslessinfrastructuresupport,andhenceisusedbyenthusiasts

l  Protectsemailatrest—buthinderssearching

l  Doesnotprotectemailheadersorothermetadata

39

Tor:TheOnionRouterl  ComputerApicksasequenceofTorrelays

(C➝E➝D)l  Distheexitnode,andpassesthetrafficto

des_na_onhostGl  Allofthesehopsareencrypted

l  BpicksrelaysF➝C➝Dl  Gcan’ttellwhichisfromAandwhichfrom

B

l  NeithercananyoneelsemonitoringG’straffic

l  ManyuseTorforanonymity:police,humanrightsworkers,spies—andcriminals(e.g.,RossUlbrichtofSilkRoadfame)

l  Mentalmodel:nested,sealedenvelopes

A D

B C

G

F

E

40

CloudCompu_ng

41

What’saCloud?l  Acloudisatradi_onalwayto

representanetwork

l  This“three-cloudnetwork”pictureisfrom1982

l  But—today“cloud”referstocompu_ngservicesprovidedviatheInternetbyanoutsideparty.

l  (Themodernusageseemstodateto1996:hAp://www.technologyreview.com/news/425970/who-coined-cloud-compu_ng/)

42

“ViatheInternet”l  Theserviceisnotprovidedon-premises

l  AnInternetlinkisnecessary

l  Thislinkprovidesanopportunityforintercep_on,lawfulorotherwise

43

“OutsideParty”l  Bydefini_on,cloudservicesareprovidedbyanoutsideparty

l  Similarinspirittothecompu_ngand_me-sharingservicebureaus,whichdatebacktothe1960s

l  Notthesameasacompany’sownremotecompu_ngfacilityl  Organiza_onscanhavea“privatecloud”,butthelegalissuesmaybeverydifferent

44

Compu_ngServicesl  Manydifferenttypesofservices

l  Storagel  Compu_ngl  Applica_onsl  Virtualmachinesl  More

45

Storagel  Diskspaceinaremoteloca_on

l  Easilyshared(andoutsidethecorporatefirewall)

l  Olenreplicatedforreliabilityl  Replicascanbeondifferentpowergrids,earthquakezones,countries,con_nents,etc.l  Datacanbemoved—ormove“byitself”—tobeclosertoitsusers

l  Expandable

l  Someoneelsecanworryaboutdiskspace,backups,security,andmore

l  Examples:Dropbox,GoogleDrive,Carbonite(forbackups),AmazonS3

l  Mentalmodel:secure,self-storagewarehouse

46

Compu_ngl  Rentcompu_ngcyclesasyouneedthem

l  Payonlyforwhatyouuse

l  Olenusedinconjunc_onwiththeprovider’scloudstorageservice

l  Examples:AmazonEC2,MicrosolAzure,GoogleCloudl  Dropboxisacloudservicethatusesadifferentprovider’scloudstorage

l  Mentalmodel:callingupatempagencyforseasonalemployees

47

Applica_onsl  Providerrunspar_cularapplica_onsforclients

l  Commontypes:websites,emailservices

l  Lesscommontypes:sharedwordprocessing,payrolls

l  Well-knownproviders:Google’sGmailandDocs,Microsol’sOutlookandOffice360,Dreamhost(webhos_ng)

l  Mentalmodel:engagingacontractorforspecifictasks

48

PlayinganAc_vePart:GoogleDocsl  Someone,usingaWebbrowser,createsadocument

l  Standardforma|ngbuAons:font,italicsorbold,copyandpaste,etc.

l  Otherswhohavetheproperauthoriza_on(some_mesjustaspecialURL)caneditthedocumentviatheirownWebbrowsers

l  Thechangesmadebyoneusershowupinreal1meinallotherusers’browserwindows

l  Inotherwords,Googleisnotjustapassiverepository;itisno_cingchangesandsendingthemoutimmediately

49

VirtualMachinesl  Normaldesktops:anopera1ngsystem(e.g.,MicrosolWindows)runsthecomputer;applica_onsrunontopoftheopera_ngsystem

l  Virtualmachines:ahypervisorrunningonasinglecomputeremulatesmul_plerealcomputers.Adifferentopera_ngsystemcanrunoneachoftheseemulatedcomputers—andeachoneisindependentoftheothersandisprotectedfromit

l  Neteffect:manycomputersthatconsumethespaceandpowerrequirementsofasinglecomputer

l  Mentalmodel:rentedofficespace

50

Loca_onofCloudServersl  Responsivenessofandeffec_vebandwidthtoaserverislimitedbyhowfarawayitisl  Theproblemisthespeedoflight—andnotevenSiliconValleycanovercomethatlimit!

l  Ittakesaminimumofaquarter-secondtosetupasecureconnec_onfromWashingtontoParis,andtwicethattoNewDelhi

l  Forperformancereasons—andindependentofpoli_calandlegalconsidera_ons—largecloudprovidersthereforeplaceservercomplexesinmanyplacesaroundtheworldl  Also:takeadvantageofcheappowerandcooling

51

WhereisDataStored?l  Modernemail:ontheserverandononeormoredevices

l  Userscan’teasilytellwhat’sontheirdevice(e.g.,phoneorlaptop)versuswhatisretrievedfromtheserverondemand

l  Itdiffersfordifferentdevicesatdifferent_mes,andmaydependontheuser’srecentac_vity

l  Whatifthedeviceandserverareindifferentjurisdic_ons?

l  (AbadfitfortheassumedbehaviormodelofStoredCommunica_onsAct)

52

SecurityandPrivacyIssuesl  Gmail:Googleapplica_onsscanemailandserveupappropriateads

l  Dropbox:usesAmazonS3foractualstorage;encryptsdatasothatAmazoncan’treadit—butDropboxcan

l  SpiderOak:dataisencryptedwiththeuser’spassword;SpiderOakcan’treadit

l  Outlook.com:blocksfileaAachmentsthatfrequentlycontainviruses

l  Many:checkpicturesforknownchildpornography

l  Many:spamfiltering

53

How the Internet Works · What is the Internet Made of? l Computers l Servers l Clients l Phones l “Things” l Routers—specialized computers that forward “packets” l Packets

Documents