YT YouTube2Text

YouTube2Text → Transcripts → 4.3 The Internet Protocol, part 1

4.3 The Internet Protocol, part 1 — Transcript

by JimKurose · 4,615 words · 663 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00[Music]
0:09in this section we're going to take a
0:11deep dive into the Internet's network
0:14layer and because there's a lot to cover
0:16here we're going to do this in two parts
0:18in the first part we're going to cover
0:20the ipv4 protocol and addressing now
0:23obviously IP is incredibly important but
0:27so too is addressing and actually
0:29address things part of ipv4 you know I
0:32used to think that addressing was sort
0:34of cut and dry and maybe even a little
0:36bit boring but I've learned that
0:38addressing has a lot to do with
0:40relationships between ISPs having to do
0:43with administrative boundaries technical
0:45issues as small and local as what
0:48happens inside a routing table lookup
0:50and hardware - as global a set of
0:52concerns as how global addressing and
0:55forwarding is done so it turns out
0:57addressing is really important and also
0:59pretty interesting in the second part
1:02we're going to cover what's known as
1:03network address translation NAT and a
1:06new version of IP known as ipv6 so I
1:09think you'll find this interesting stay
1:11tuned so let's start with a big picture
1:14view of the Internet's network layer
1:16remember we learned earlier about the
1:18network layers control and data plane in
1:21the forwarding table that's the router
1:23local table that determines the output
1:25port to which a router will forward an
1:28incoming data gram and we also learned
1:30in the introduction to the network layer
1:32that the forwarding tables contents
1:34determined either by distributed routing
1:37protocols or by a router external Sdn
1:40controller will cover routing algorithms
1:42and Sdn controllers when we study the
1:45network layers control plane a key part
1:48of the internets network layer is of
1:50course the IP protocol the celebrated IP
1:53protocol but what exactly is the IP
1:55protocol well by looking at this diagram
1:58you can see that the IP protocol is not
2:00about routing algorithms or Sdn
2:03controllers those are control plane
2:05functions instead the IP protocols about
2:08the data gram format it's about how IP
2:11addresses are structured and interpreted
2:14and it's about packet handling
2:16conventions like how to fragment large
2:18packets into smaller packets so IP is
2:22really just a piece albeit a really
2:24important piece of the Internet's
2:26network layer the ICMP protocol which
2:30we'll cover as part of the network layer
2:31control plane is also part of the
2:34network layer since the IP protocol is
2:37about data Graham formats addressing and
2:40packet handling conventions let's then
2:42start our study of IP by looking at the
2:45IP Datagram format using this figure
2:48we'll walk through the fields of the IP
2:50Datagram now I know this might seem a
2:52little bit dry but really anyone who
2:55takes a networking course really needs
2:57to know this sort of like eating your
2:59vegetables you know it's good for you
3:00and so you really do have to do it and
3:02hopefully you'll even love it or find it
3:06interesting anyway well first there's
3:09the version number field these 4 bits
3:11specify the IP protocol version of the
3:14data Graham here of course we're looking
3:16at the ipv4 header because an ipv4
3:20Datagram can contain a variable number
3:22of options which we see here in the data
3:25gram header the header length field
3:28indicates how many bytes are in the
3:30header this lets the host or router know
3:32where the payload actually begins in the
3:34IP Datagram most IP datagrams don't
3:37contain options
3:38so the typical IP Datagram header goes
3:40from here to here a 20-byte header the
3:43data gram length field indicates the
3:46total number of bytes in the IP Datagram
3:48header plus payload this is needed
3:50because the payload can also be a
3:52variable size since this fields 16 bits
3:55long the theoretical maximum size of an
3:59IP Datagram is 64 K bytes but data grams
4:02are usually no larger than 1,500 bytes
4:05which allows a data gram to fit nicely
4:07inside the payload field of a maximally
4:10sized Ethernet link layer frame the type
4:13of service bits were included in the
4:15ipv4 header to allow different types of
4:18data grams to be distinguished from each
4:20other the definition and use of these
4:22bits is really evolved over time for our
4:25purposes the most important type of
4:27service bits are the two bits used for
4:30explicit congestion notification ecn and
4:32remember that's a topic we study
4:35in detail in the transport layer a
4:37router sets these two bits to indicate
4:39congestion the remaining six types of
4:42service bits are used to differentiate
4:44among different classes of traffic which
4:46can then be given different service
4:48based on their traffic class using the
4:50buffering and scheduling algorithms we
4:53just studied in the previous section the
4:55TTL field is a counter that decremented
4:59by 1 each time a Datagram passes through
5:01a router if the TTL count ever reaches 0
5:04the data gram must be dropped at the
5:06router and so the TTL field is used to
5:09ensure that packets don't loop forever
5:11or even for a long time if there are
5:14forwarding loops the upper-layer
5:16protocol field indicates the transport
5:18layer protocol to which the payload in
5:20this IP Datagram is going to be passed
5:23for example a value of 6 indicates that
5:26this data gram contains a TCP segment
5:28that will be passed up to TCP a value of
5:3117 indicates that the payload is a UDP
5:33segment the IP ID field flags and
5:37fragment offset field are used when a
5:40single large IP datagrams fragmented
5:43into multiple smaller data grams this
5:46doesn't happen often and actually these
5:48fields don't even show up in IP version
5:506 and so we won't see anything more
5:52about them here you can read about them
5:54more by following the pointers in our
5:57textbook the header checksum is an
6:00internet checksum computed over the
6:02contents of the IP header remember we
6:05studied the internet checksum earlier
6:07and note that since some of the IP
6:09header fields change each time a data
6:12gram passes through a router the TTL
6:14field is decremented for example the
6:16checksum needs to be recomputed at each
6:18router that the data gram passes through
6:21that can be time-consuming and may be as
6:24a result of this the header checksum
6:26field was also removed in ipv6 we've
6:29already talked about the 32-bit source
6:32and destination IP address fields
6:34remember that we just saw in destination
6:36based forwarding that an IP router uses
6:39the data grams destination IP address
6:41here to look up the appropriate router
6:44output port to which the data gram
6:46should be
6:47forward it and then finally there are a
6:50number of optional fields in the data
6:52gram header we'll skip those here and
6:54then of course the data grams payload
6:56itself the payloads the data transport
6:59layer segment that IP will deliver up to
7:02the transport layer protocol whoa
7:05well that was a lot of fields to think
7:07about and to go through and as we
7:09mentioned ipv6 is a lot more streamlined
7:12than ipv4 and so we'll find fewer header
7:15fields there let's start our discussion
7:20of IP addressing with a few basic
7:22concepts the first important thing to
7:24know is that an IP address does not
7:27identify a host or router per se but
7:30rather an interface a link layer
7:33interface on a host or router routers
7:36pretty much always have more than one
7:38interface that is to say multiple
7:40incoming and outgoing links and hosts
7:43often do as well for example my laptop
7:45has a wired Ethernet interface and a
7:48wireless 802 11 interface each with a
7:51different IP address as we've seen 32
7:55bit IP addresses are written in so
7:58called dotted decimal notation with each
8:00decimal number corresponding to an 8-bit
8:03byte in the address field here the
8:0632-bit binary address is to 23.1 1.1 in
8:11dotted decimal notation well you know
8:15looking at this diagram it might be a
8:16little bit unsatisfying to you it looks
8:19like the link layer interfaces these
8:21black lines just sort of vaguely end up
8:24in a blue cloud what's up with that how
8:27are these interfaces actually connected
8:29together so that one interface can
8:31communicate directly with another
8:33interface that's attached to the same
8:35blue cloud well remember that we're at
8:38the network layer right now and the
8:40connection of these interfaces is done
8:42via link layer technologies which are
8:45the layer below what we're studying
8:46right now so one answer to your question
8:49would be well be patient we'll get to
8:51that but if you're just dying to see
8:53what the complete picture looks like
8:55here's what it is
8:57the host and the router interfaces might
8:59be connected through wired Ethernet
9:00switches or they might be connected
9:02through a NATO 2.11 wireless network but
9:05there will be a link layer local area
9:08network or point-to-point link protocol
9:10typically Ethernet or Wi-Fi in practice
9:13that will be used to connect the
9:15interfaces and we really will learn
9:18about this just a bit later so stay
9:20tuned and for now back to the network
9:22layer if you take a close look at this
9:26figure in particular the IP addresses in
9:29this figure you'll notice that the
9:30interfaces that are connected to each
9:32other have similar but not identical IP
9:35addresses that's because they all belong
9:38to the same subnet so what is a subnet
9:41well each subnet is a piece of the
9:44network that contains all devices that
9:46can reach each other without passing
9:49through an intervening layer three that
9:51is to say network layer router that is
9:54they're directly connected to each other
9:56via some link layer technology and
9:59here's how that ties into addressing an
10:01IP address has two parts a subnet part
10:04and a host part and if two interfaces
10:07are on the same subnet their IP
10:10addresses must have a common subnet part
10:13and of course they'll have different
10:14host parts so let's be very specific in
10:18defining subnets to find a subnet
10:21attached each interface from its host or
10:24router as we've done here and this will
10:27leave islands of isolated networks and
10:30so in this example here we have three
10:32subnets now let's look at the addressing
10:35remember that interfaces on the same
10:38subnet will have the same subnet part of
10:41their IP addresses the so-called subnet
10:44mask says how many high order bits they
10:46have in common let's assume that's 24
10:49bits in this case this lower subnet here
10:52has a 24-bit address of 220 3.1.3 we say
10:58220 3.1 3/24 and these three interfaces
11:05have addresses that all start with 223
11:09dot
11:09three the next subnet is two 23.1 1/24
11:16note that these four interfaces have
11:19addresses that start with two twenty
11:21three dot one dot one and this third
11:24subnet address is two twenty three dot
11:27one dot 2/24 let's take a look at this
11:32network with three routers and seven
11:34hosts can you find the subnets think
11:37about it well let's disconnect the
11:41interfaces from their host or router to
11:43see the subnets and here are the slash
11:46twenty four subnet addresses this slash
11:50notation that we've been using has a
11:52name it's called classless inter-domain
11:54routing CIDR and actually pronounced
11:57cider due to its historical origins the
12:01subnet portion of the address can be any
12:03number of bits x and a cider i's address
12:06is one of the form ABCD / X where X is
12:12the number of bits in the subnet address
12:13and sometimes the D is emitted when it's
12:16zero here's an example of a 32 bit IP
12:19address and its side Erised
12:21dotted decimal notation well let's wrap
12:25up our discussion of ipv4 and addressing
12:29by posing a question that maybe you've
12:31been wondering about how does one
12:32actually get an IP address in the first
12:35place and really that's two questions
12:37how does a host get an address within
12:40the subnet range associated with the
12:42subnet that it's going to be joining and
12:44then sort of the bigger picture question
12:46how does a network how does a sub
12:48network get a range of addresses that
12:50are going to be used by devices by
12:53interfaces within that sub network let's
12:55look at both of these questions well
12:59let's start with the first question how
13:01does a host get an IP address for its
13:03interface well back in the day a
13:06sysadmin used to actually manually edit
13:09the IP address into a file on that host
13:12but we've progressed a lot since then
13:14using a plug and play or what sometimes
13:17known as a zero Kampf zero configuration
13:20approach that's to say an approach
13:22that uses a protocol between a host and
13:25a server rather than manual
13:27configuration to get an IP address the
13:30need for a zeroconf approach is
13:31hopefully obvious with billions of hosts
13:34worldwide more than half being mobile
13:36and these are hosts that are constantly
13:38attaching to a network and then later
13:40detaching and then attaching over and
13:42over again the protocol used by a host
13:45to get an IP address on a network is
13:47known as DHCP the dynamic host
13:50configuration protocol in order to
13:54allocate a plug-and-play IP address
13:57using DHCP a network will need to have a
14:00DHCP server to perform that function and
14:03an arriving client that is a host
14:05wanting an IP address will request and
14:08receive an IP address from the DHCP
14:10server using the DHCP protocol when a
14:15host leaves the network it'll give up
14:17its IP address which can then later be
14:19reused by another host or reclaimed or
14:22may be renewed by this host at a later
14:25point in time we're going to take a look
14:27at for DHCP messages that we'll want to
14:30study there are more but these are
14:32really the key ones discover offer
14:35request in acknowledgment let's see how
14:37DHCP works here's the scenario we'll
14:42look at using the simple network that
14:44we've seen before and in this network we
14:46see the DHCP server here well typically
14:50a DHCP server would be co-located in a
14:53router and serving all of the subnets to
14:56which that router is attached but we've
14:58shown it here as a separate server which
15:00also happens with an IP address of 223
15:0512.5 and here's the arriving client that
15:08will want an IP address in the 223 one
15:132/24 subnet now let's take a look at
15:16what the DHCP clients server message
15:19exchange actually looks like
15:22in the first step the arriving client
15:24broadcasts a DHCP message that will be
15:27received on the interfaces of all of the
15:30hosts and routers in the subnet to which
15:33it's attaching the discover message
15:35basically says hey is there a DHCP
15:38server out there this is a form of
15:40service discovery the host knows the
15:43service it needs DHCP so it sends out a
15:46broadcast to discover the server that
15:48can provide DHCP service DHCP runs over
15:52UDP the client uses port 68 and the
15:56server will use port 67 in particular
15:59the server will be listening on a socket
16:01associated with UDP port 67 for incoming
16:06DHCP messages and here are the details
16:09of the fields in the DHCP discover
16:11message the clients source IP address
16:14containing the UDP segment that contains
16:17the DHCP discover message is zero
16:19because the client doesn't have an IP
16:21address yet and the destination IP
16:24address is the IP broadcast address that
16:28is all ones 255 255 255 255 and the UDP
16:34destination port number is 67 as we just
16:38noted before hopefully there'll be a
16:40DHCP server sitting out there somewhere
16:43on the subnet listening for this service
16:45discovery message note also that there's
16:47a transaction ID field it's got a value
16:50654 here and this will be used by the
16:53client to match any reply to this
16:56request here's the second step any DHCP
17:02server that receives this broadcast
17:04Discoverer message and there might be
17:06several such servers can reply with a
17:09DHCP offer message this message
17:12basically says hey I'm a DHCP server and
17:15here's an IP address that you can use if
17:18we look at the details of this offer
17:20message we see that it comes from the
17:23DHCP server with an IP address of 223
17:2612.5 as in our figure earlier and from
17:30port number 67 this offer message is
17:33being broadcast to all interfaces
17:35on the subnet the DHCP message contains
17:38the IP address that the requesting host
17:41can use that's the yia ddr your internet
17:46address field and a lifetime in this
17:49case of 3600 seconds that the address
17:52can be used for note that the
17:54transaction ID here matches the
17:56transaction ID of the initial offer
17:58message that's to say this is a reply to
18:02that message a client can receive offers
18:04from multiple DHCP servers and this
18:07might occur for example if there are
18:09multiple routers on that subnet now the
18:12first two steps that we've seen here are
18:14actually optional the way to think of
18:16the third step which could also be the
18:18first step if the two optional steps
18:20aren't taken is that the arriving client
18:22comes in and says hey here's an IP
18:25address that I want to use maybe that's
18:28the IP address that it was told it could
18:30use and step two if that step was
18:32executed or maybe it's an address that
18:35it already has in which case the client
18:37is really renewing its use of that
18:39address or perhaps it's an address that
18:42the clients used before you can see here
18:45that this message is also broadcast and
18:47then it contains the IP address that the
18:50host is proposing to use and that
18:52there's a new transaction ID and a
18:54lifetime as well the final message is
18:57the ACK message from the DHCP server
18:59saying basically okay you can use that
19:02IP address with this given lifetime now
19:06it turns out that a host needs more
19:08configuration parameters than just an IP
19:10address in order to be able to function
19:12in particular it also needs to know the
19:15IP address of the first hop router since
19:18any outgoing packet from the client will
19:20need to be sent to that first hop router
19:22it might also need the address of a DNS
19:24server to use and also need the netmask
19:27the number of bits and its IP address
19:29belonging to the subnet well this could
19:32all be configured manually but it can
19:35also be optionally specified in a DHCP
19:38message and this is usually the case so
19:41the other really big question that we
19:43still need to ask is how does a network
19:45get a range of addresses to assign to
19:48the devices
19:49actually to the device interfaces in its
19:52network well in many cases it will be
19:55allocated a range of addresses in the IP
19:58address range that belongs to the ISP to
20:02which it is a customer network in this
20:05example here the higher-level ISP has a
20:08slash 20 address range this is because
20:11then for example divide its address
20:13space up into 8 say slash 23 address
20:17ranges as shown here and allocate each
20:20of these out to one of its eight client
20:23networks and now we can start to see a
20:27really critical time between addressing
20:30and routing something that we'll come
20:32back to again when we get to the network
20:34layer control plane in this example the
20:37parent ISP which we'll call fly-by-night
20:40ISP in this example has eight client
20:43ISPs as shown in the figure here
20:46fly-by-night need only advertise one
20:49address prefix 223 16/20 to the rest of
20:55the global Internet
20:56this single advertised address prefix
20:59will be sufficient for the rest of the
21:01internet to be able to route to the two
21:04to the twelfth addresses in this is P is
21:07address range this is an example of
21:10what's called address aggregation
21:12sometimes also called route aggregation
21:14or route summarization note that in this
21:17figure the second is P let's call it is
21:20peas our US says hey send me anything in
21:24the address range 190 9.30 1/16 we'll
21:29come back to this in just a second
21:31well this all looks very neat and
21:34orderly but in life well think just
21:37generally aren't so neat or ever so
21:39perfectly hierarchical in particular
21:42let's ask ourselves what would happen
21:44hypothetically if one of the eight
21:46client ISPs say organization one wants
21:50to change ISPs from fly-by-night to ISPs
21:54our us while keeping its range of
21:57network addresses 223
22:0118/20 three well fly-by-night continues
22:05to say send me anything with address
22:08prefix 223 16/20 as before and as we can
22:13see here but here's the critical thing
22:16now is peas are us says well send me
22:19anything with address range 190 9.30
22:231/16 as before but now it also says also
22:28send me anything with address range 200
22:32dot 23.18.38
22:39and note that 223 18 / 23 is contained
22:44in fly-by-nights advertised 223 dot
22:5016/20 address range so what are we to do
22:54here
22:54well what's absolutely critical here is
22:57that ISPs our US has advertised a longer
23:0223 bit prefix 223 18 / 23 a more
23:08specific prefix then the 20 bit prefix
23:12223 slash 16/20 advertised by
23:16fly-by-night and i hope oh yeah light
23:20bulb has just gone off in your head
23:22remember when we were studying forward
23:24tables when we were looking at what's
23:26inside a router when looking up a 32 bit
23:29IP address the matching forwarding table
23:32entry is that with the longest prefix
23:34match that's precisely why packets with
23:37addresses and organization ones address
23:40range will be forwarded to ISPs our US
23:44which has advertised the longer prefix
23:47them
23:48well we've now seen how address
23:51allocation forwarding table lookup using
23:54longest prefix matching in the BGP
23:57routing protocol that disseminates
23:59prefixes are all really intimately tied
24:02together and if you've got all of this
24:04you've really mastered and tied together
24:06a lot of key concepts if it's still not
24:10100% clear we'll come back to this again
24:12when we study the control plane but
24:15maybe think a little bit more about it
24:16now if you don't have it quite down so
24:20we've now seen how a host gets an IP
24:23address from a DHCP server and we've
24:26seen how customer ISPs can get an
24:28address range from their provider ISP
24:30but we still haven't answered the basic
24:33question at the highest level how does
24:36an ISP get a block of addresses and the
24:39answer to this final question is that
24:41the ipv4 address space is owned
24:44administered and allocated by an entity
24:47known as the Internet Corporation for
24:49Assigned Names and numbers I can I can
24:53in turn allocates addresses out to five
24:55regional registries which then allocate
24:58out their portions of the address space
25:01to ISPs I can also performs a number of
25:05other critical internet functions like
25:07managing the route DNS servers managing
25:11domain names and allocating protocol
25:13numbers remember when we study the ipv4
25:16data gram format and said that the upper
25:19layer protocol number of six corresponds
25:22to TCP and that the number 17
25:24corresponds to UDP those numbers are
25:27also controlled by ICANN well let's wrap
25:30up here by considering ipv4 32-bit
25:34address space in 2011 ICANN central
25:38allocated out the last of its available
25:40chunks of 32-bit address space to the
25:44regional registries ICANN has no more
25:46address space to allocate out although
25:49there's some unused address space still
25:51sitting in the regional registries in
25:53the next section we'll look at network
25:55address translation a technique that
25:57allows a single IP address to be used by
26:00multiple IP devices and will also look
26:04at IP version 6 which has a much larger
26:07128-bit address space ipv6 was begun in
26:10the 1990s specifically because the IETF
26:14foresaw the exhaustion of the ipv4
26:17address space well you see this quote
26:20here from Vint Cerf who along with Bob
26:23Kahn we can really credit as being one
26:25of the two founding fathers of the
26:26Internet
26:27if we had to choose just two they've
26:29both been asked a lot about the early
26:31days of the experimental ARPANET that
26:34later grew into today's Internet and
26:36about the 32-bit ipv4 address space in
26:39particular and I love this quote from
26:43Vic who the hell knew how much address
26:45space we needed
26:46remember when today's internet
26:48architecture was being defined in the
26:501970s it was a relatively small
26:52Department of Defense project so let's
26:57leave the last words here to be from
26:59Vint Cerf I heard him give a really
27:01great sort of fireside chat type of talk
27:04at Georgetown University in 2017 when I
27:07was working in DC and he talked about a
27:09lot of really interesting topics but one
27:12of them had to do with IPS 32-bit
27:15address space so let's hear it from Vint
27:18how did we actually end up with a 32-bit
27:20address space what calculations did they
27:23go through take a look so now we have
27:25four different kinds of networks the
27:27Ethernet the packet main unit the packet
27:29satellite net in the urban did the
27:30original ARPANET so we sat down and
27:33tried to figure out how are we going to
27:35do that and after about six months we
27:38had designed the basic protocols that
27:40you use today it was just TCP later and
27:44we split off the internet protocol for
27:46real-time communication so the question
27:49then was you know how big did we expect
27:51this to be well honestly we sat down and
27:54said this is going to be used by the
27:55Defense Department it has to work
27:57everywhere in the world because you
27:58never know where the Defense Department
28:00is going to have to operate so we said
28:03okay how big is this going to be it's
28:05got to be global how many networks will
28:07there be per country and we said well we
28:11had finished the ARPANET which is Anand
28:14national scale network so we thought
28:16well maybe they'll be 2 per country
28:18so let me sum competition so we said ok
28:21- how many countries are there and there
28:24wasn't any Google to ask back then so we
28:27guessed it 128 cuz that's a power of 2
28:30and that's a programmers thinking so 2
28:32times 128 is 256 networks and that's 8
28:36bits of identifier and then we said how
28:39many computers will there be per network
28:41and we said remember this is 1973 these
28:45are giant air conditioned computers you
28:48know that serve a thousand people in
28:50time shared modes so we're saying how
28:52about 16 million you know what the heck
28:55so 16 million is 24 bits so we ended up
28:59with a 32-bit address space and if you
29:01do the math it's 4.3 billion possible
29:04terminations for the network in 1973 so
29:07the answer that was more than there were
29:09people in the world and they think so we
29:11thought well that ought to be enough to
29:12do this experiment right so that's what
29:16we used so people say if you had it to
29:20do over again you know what would you do
29:22and one answer is well I guess I would
29:25have chosen 128 bit address space and we
29:27don't have to go through this painful
29:29transition on the other hand can you
29:31imagine making it an argument of the
29:34form I need 3.4 times 10 to the 38th
29:37addresses in order to do this experiment
29:39how many networks do you have now three
29:44how many computers do you have on the
29:46networks 500 he said something doesn't
29:50quite add up
29:51does it so I didn't do it because that
29:54was it wouldn't pass the red face test
29:55but I wish I could have

About this transcript

This page contains the full transcript of 4.3 The Internet Protocol, part 1 by JimKurose, generated from the public captions YouTube serves with the video. The transcript has 4,615 words across 663 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.