Introduction                                   to                         the Internet Protocols                           C                       R                               C       S                   Computer Science Facilities Group                               C       I                       L                       S                                 RUTGERS                   The State University of New Jersey                                 3 July 1987 This is an introduction to the Internet networking protocols (TCP/IP). It  includes  a  summary  of  the  facilities  available   and   brief descriptions of the major protocols in the family. Copyright  (C)  1987,  Charles  L. Hedrick.  Anyone may reproduce this document, in whole or in  part,  provided  that:    (1)  any  copy  or republication  of  the entire document must show Rutgers University as the source, and must include this notice; and (2)  any  other  use  of this  material  must reference this manual and Rutgers University, and the fact that the material is copyright by Charles Hedrick and is used by permission.   Unix is a trademark of AT&T Technologies, Inc.                             Table of Contents     1. What is TCP/IP?                                                1    2. General description of the TCP/IP protocols                    5        2.1 The TCP level                                             7        2.2 The IP level                                             10        2.3 The Ethernet level                                       11    3. Well-known sockets and the applications layer                 12        3.1 An example application: SMTP                             15    4. Protocols other than TCP: UDP and ICMP                        17    5. Keeping track of names and information: the domain system     18    6. Routing                                                       20    7. Details about Internet addresses: subnets and broadcasting    21    8. Datagram fragmentation and reassembly                         23    9. Ethernet encapsulation: ARP                                   24    10. Getting more information                                     25                                                                        i   This document is a brief introduction to TCP/IP, followed by advice on what to read for more information.  This  is  not  intended  to  be  a complete  description.    It  can  give  you  a reasonable idea of the capabilities of the protocols.  But if you need to know any details of the  technology,  you  will  want  to  read  the  standards  yourself. Throughout the text, you will find references to the standards, in the form of "RFC" or "IEN" numbers.  These are document numbers. The final section of this  document  tells  you  how  to  get  copies  of  those standards.   1. What is TCP/IP?  TCP/IP  is a set of protocols developed to allow cooperating computers to share resources across a network.  It was developed by a  community of  researchers centered around the ARPAnet.  Certainly the ARPAnet is the best-known TCP/IP network.  However as of June, 87, at  least  130 different  vendors  had products that support TCP/IP, and thousands of networks of all kinds use it. First some basic definitions.  The most accurate name for the  set  of protocols we are describing is the "Internet protocol suite".  TCP and IP are two of the protocols in this suite.  (They  will  be  described below.)    Because  TCP and IP are the best known of the protocols, it has become common to use the term TCP/IP or IP/TCP  to  refer  to  the whole  family.  It is probably not worth fighting this habit.  However this can lead to some oddities.  For example, I  find  myself  talking about  NFS as being based on TCP/IP, even though it doesn't use TCP at all.  (It does use IP.  But it  uses  an  alternative  protocol,  UDP, instead  of TCP.  All of this alphabet soup will be unscrambled in the following pages.) The Internet is a  collection  of  networks,  including  the  Arpanet, NSFnet, regional networks such as NYsernet, local networks at a number of University and research institutions,  and  a  number  of  military networks.  The term "Internet" applies to this entire set of networks. The subset of them that is managed by the  Department  of  Defense  is referred  to  as the "DDN" (Defense Data Network).  This includes some research-oriented networks, such as  the  Arpanet,  as  well  as  more strictly  military  ones.    (Because much of the funding for Internet protocol developments is done via  the  DDN  organization,  the  terms Internet  and  DDN  can  sometimes  seem  equivalent.)    All of these networks are connected to each other.  Users can  send  messages  from any  of  them  to  any other, except where there are security or other policy restrictions on access.    Officially  speaking,  the  Internet protocol  documents  are  simply  standards  adopted  by  the Internet community for its own use.  More recently, the Department  of  Defense issued a MILSPEC definition of TCP/IP.  This was intended to be a more formal definition, appropriate for use in  purchasing  specifications. However  most  of  the  TCP/IP community continues to use the Internet standards.  The MILSPEC version is intended to be consistent with it. Whatever it is called, TCP/IP is a family of protocols.  A few provide                                   1   "low-level" functions needed for many applications.  These include IP, TCP, and UDP.  (These will be described in a bit more  detail  later.) Others are protocols for doing specific tasks, e.g. transferring files between computers, sending mail, or finding out who is  logged  in  on another   computer.      Initially  TCP/IP  was  used  mostly  between minicomputers or mainframes.  These machines had their own disks,  and generally  were self-contained.  Thus the most important "traditional" TCP/IP services are:    - file transfer.  The file transfer protocol (FTP) allows a user on      any computer to get files from another computer, or to send files      to another computer.  Security is handled by requiring  the  user      to  specify  a  user  name  and  password for the other computer.      Provisions are made for handling file transfer  between  machines      with different character set, end of line conventions, etc.  This      is not quite the same thing as more recent "network file  system"      or  "netbios"  protocols, which will be described below.  Rather,      FTP is a utility that you run any time you want to access a  file      on  another  system.    You  use  it to copy the file to your own      system.  You then work with the local copy.   (See  RFC  959  for      specifications for FTP.)    - remote  login.    The network terminal protocol (TELNET) allows a      user to log in on any other computer on the network.  You start a      remote session by specifying a computer to connect to.  From that      time until you finish the session, anything you type is  sent  to      the  other  computer.   Note that you are really still talking to      your own computer.  But the telnet program effectively makes your      computer invisible while it is running.  Every character you type      is sent directly to the other system.  Generally, the  connection      to  the  remote  computer  behaves much like a dialup connection.      That is, the remote system will ask you to  log  in  and  give  a      password, in whatever manner it would normally ask a user who had      just dialed it up.  When you log off of the other  computer,  the      telnet  program exits, and you will find yourself talking to your      own computer.  Microcomputer implementations of telnet  generally      include  a  terminal  emulator  for some common type of terminal.      (See RFC's 854 and 855 for specifications for  telnet.    By  the      way,  the  telnet protocol should not be confused with Telenet, a      vendor of commercial network services.)    - computer mail.  This allows you to  send  messages  to  users  on      other  computers.    Originally, people tended to use only one or      two specific computers.  They  would  maintain  "mail  files"  on      those machines.  The computer mail system is simply a way for you      to add a message to another user's mail file.    There  are  some      problems  with  this  in  an environment where microcomputers are      used.  The most serious is that a micro is  not  well  suited  to      receive  computer  mail.    When you send mail, the mail software      expects to be able  to  open  a  connection  to  the  addressee's      computer, in order to send the mail.  If this is a microcomputer,      it may be turned off, or it may be running an  application  other      than  the mail system.  For this reason, mail is normally handled      by a larger system, where it is practical to have a  mail  server      running all the time.  Microcomputer mail software then becomes a                                   2        user interface that retrieves mail from the mail  server.    (See      RFC  821  and  822 for specifications for computer mail.  See RFC      937 for a protocol designed for microcomputers to use in  reading      mail from a mail server.)   These  services  should  be  present  in any implementation of TCP/IP, except that micro-oriented implementations may  not  support  computer mail.  These traditional applications still play a very important role in TCP/IP-based networks.  However more recently,  the  way  in  which networks  are  used has been changing.  The older model of a number of large, self-sufficient computers is beginning to  change.    Now  many installations    have    several   kinds   of   computers,   including microcomputers, workstations, minicomputers, and  mainframes.    These computers  are  likely  to be configured to perform specialized tasks. Although people are still likely to work with one  specific  computer, that  computer  will  call on other systems on the net for specialized services.  This has  led  to  the  "server/client"  model  of  network services.    A server is a system that provides a specific service for the rest of the network.  A client is another system  that  uses  that service.    (Note  that the server and client need not be on different computers.  They could be  different  programs  running  on  the  same computer.)    Here  are  the  kinds  of servers typically present in a modern computer setup.  Note that these computer services can  all  be provided within the framework of TCP/IP.    - network  file  systems.   This allows a system to access files on      another computer in a somewhat more  closely  integrated  fashion      than FTP.  A network file system provides the illusion that disks      or other devices from one system are directly connected to  other      systems.    There  is no need to use a special network utility to      access a file on another system.  Your computer simply thinks  it      has  some  extra disk drives.  These extra "virtual" drives refer      to the other system's disks.    This  capability  is  useful  for      several different purposes.  It lets you put large disks on a few      computers, but still give others access to the disk space.  Aside      from the obvious economic benefits, this allows people working on      several computers  to  share  common  files.    It  makes  system      maintenance  and  backup  easier, because you don't have to worry      about updating  and  backing  up  copies  on  lots  of  different      machines.    A  number  of  vendors  now  offer  high-performance      diskless computers.  These computers have no disk drives at  all.      They  are  entirely dependent upon disks attached to common "file      servers".   (See  RFC's  1001  and  1002  for  a  description  of      PC-oriented   NetBIOS   over   TCP.     In  the  workstation  and      minicomputer area, Sun's Network File System is more likely to be      used.    Protocol  specifications  for  it are available from Sun      Microsystems.)    - remote printing.  This allows you to  access  printers  on  other      computers  as if they were directly attached to yours.  (The most      commonly used protocol is the remote  lineprinter  protocol  from      Berkeley  Unix.  Unfortunately, there is no protocol document for      this.  However the C code is easily obtained  from  Berkeley,  so      implementations are common.)                                   3      - remote  execution.   This allows you to request that a particular      program be run on a different computer.  This is useful when  you      can  do  most  of  your work on a small computer, but a few tasks      require the resources of a larger system.  There are a number  of      different  kinds  of remote execution.  Some operate on a command      by command basis.  That is, you request that a  specific  command      or  set  of commands should run on some specific computer.  (More      sophisticated versions will choose a system that  happens  to  be      free.)    However  there are also "remote procedure call" systems      that allow a program to  call  a  subroutine  that  will  run  on      another  computer.    (There  are  many  protocols  of this sort.      Berkeley Unix contains two servers to execute commands  remotely:      rsh  and  rexec.   The man pages describe the protocols that they      use.  The user-contributed software with Berkeley 4.3 contains  a      "distributed  shell"  that  will  distribute tasks among a set of      systems, depending upon load.  Remote procedure  call  mechanisms      have  been  a  topic  for research for a number of years, so many      organizations have implementations of such facilities.  The  most      widespread commercially-supported remote procedure call protocols      seem to be Xerox's Courier and Sun's RPC.  Protocol documents are      available  from  Xerox and Sun.  There is a public implementation      of Courier over TCP as part of the user-contributed software with      Berkeley  4.3.   An implementation of RPC was posted to Usenet by      Sun, and also appears as part of  the  user-contributed  software      with Berkeley 4.3.)    - name  servers.    In  large  installations, there are a number of      different collections of names that have to  be  managed.    This      includes  users  and their passwords, names and network addresses      for computers, and accounts.  It becomes  very  tedious  to  keep      this data up to date on all of the computers.  Thus the databases      are kept on a small number of systems.  Other systems access  the      data over the network.  (RFC 822 and 823 describe the name server      protocol used to keep track of host names and Internet  addresses      on  the  Internet.    This  is  now a required part of any TCP/IP      implementation.  IEN 116 describes an older name server  protocol      that is used by a few terminal servers and other products to look      up host names.  Sun's  Yellow  Pages  system  is  designed  as  a      general  mechanism to handle user names, file sharing groups, and      other databases commonly used by Unix  systems.    It  is  widely      available  commercially.    Its  protocol definition is available      from Sun.)    - terminal servers.  Many installations no longer connect terminals      directly  to  computers.    Instead they connect them to terminal      servers.  A terminal server is simply a small computer that  only      knows  how  to  run  telnet  (or some other protocol to do remote      login).  If your terminal is  connected  to  one  of  these,  you      simply  type the name of a computer, and you are connected to it.      Generally it is possible to have active connections to more  than      one  computer  at  the  same time.  The terminal server will have      provisions to switch between connections rapidly, and  to  notify      you  when  output  is  waiting for another connection.  (Terminal      servers use the telnet protocol, already mentioned.  However  any      real terminal server will also have to support name service and a                                   4        number of other protocols.)    - network-oriented  window  systems.      Until   recently,   high-      performance  graphics  programs had to execute on a computer that      had  a  bit-mapped  graphics  screen  directly  attached  to  it.      Network  window  systems  allow  a  program to use a display on a      different computer.  Full-scale network window systems provide an      interface  that  lets you distribute jobs to the systems that are      best  suited  to  handle  them,  but  still  give  you  a  single      graphically-based  user  interface.  (The most widely-implemented      window system is X. A  protocol  description  is  available  from      MIT's  Project  Athena.  A reference implementation is publically      available from MIT.  A number  of  vendors  are  also  supporting      NeWS,  a window system defined by Sun.  Both of these systems are      designed to use TCP/IP.)   Note that some of the  protocols  described  above  were  designed  by Berkeley,  Sun,  or other organizations.  Thus they are not officially part of the Internet protocol suite.   However  they  are  implemented using  TCP/IP, just as normal TCP/IP application protocols are.  Since the protocol definitions are not  considered  proprietary,  and  since commercially-support  implementations  are  widely  available,  it  is reasonable to think of these protocols as being  effectively  part  of the  Internet  suite.   Note that the list above is simply a sample of the sort of services  available  through  TCP/IP.    However  it  does contain   the  majority  of  the  "major"  applications.    The  other commonly-used protocols tend to be specialized facilities for  getting information  of  various  kinds, such as who is logged in, the time of day, etc.  However if you need a facility that is not listed here,  we encourage  you  to  look  through  the  current  edition  of  Internet Protocols (currently RFC 1011),  which  lists  all  of  the  available protocols,   and   also   to   look   at  some  of  the  major  TCP/IP implementations to see what various vendors have added.   2. General description of the TCP/IP protocols  TCP/IP is a layered set of protocols.  In  order  to  understand  what this  means,  it is useful to look at an example.  A typical situation is sending mail.  First, there is a protocol for mail.  This defines a set  of  commands which one machine sends to another, e.g. commands to specify who the sender of the message is, who it is being sent to, and then  the  text  of  the  message.  However this protocol assumes that there is a way to communicate  reliably  between  the  two  computers. Mail,  like  other  application  protocols,  simply  defines  a set of commands and messages to be sent.  It is designed to be used  together with  TCP and IP. TCP is responsible for making sure that the commands get through to the other end.  It keeps track of  what  is  sent,  and retransmitts anything that did not get through.  If any message is too large for one datagram, e.g. the text of the mail, TCP will  split  it up  into  several  datagrams,  and  make  sure  that  they  all arrive correctly.  Since these functions are needed  for  many  applications, they are put together into a separate protocol, rather than being part                                   5   of the specifications for sending mail.   You  can  think  of  TCP  as forming a library of routines that applications can use when they need reliable network communications with another computer.  Similarly, TCP calls  on the services of IP.  Although the services that TCP supplies are needed by  many  applications,  there  are  still  some  kinds  of applications  that  don't  need them.  However there are some services that every application needs.  So these services are put together into IP.    As  with TCP, you can think of IP as a library of routines that TCP calls on, but which is also available to applications  that  don't use  TCP.    This  strategy  of building several levels of protocol is called "layering".  We think of  the  applications  programs  such  as mail,  TCP, and IP, as being separate "layers", each of which calls on the services of the layer below it.   Generally,  TCP/IP  applications use 4 layers:    - an application protocol such as mail    - a  protocol  such  as  TCP  that  provides  services need by many      applications    - IP, which provides the basic  service  of  getting  datagrams  to      their destination    - the  protocols  needed to manage a specific physical medium, such      as Ethernet or a point to point line.   TCP/IP is based on the "catenet model".  (This is  described  in  more detail  in  IEN 48.)  This model assumes that there are a large number of independent networks connected together  by  gateways.    The  user should  be able to access computers or other resources on any of these networks.   Datagrams  will  often  pass  through  a  dozen  different networks  before  getting  to  their  final  destination.  The routing needed to accomplish this should be completely invisible to the  user. As  far  as  the  user  is concerned, all he needs to know in order to access another system is an "Internet address".  This  is  an  address that looks like 128.6.4.194.  It is actually a 32-bit number.  However it is normally written as 4 decimal numbers, each representing 8  bits of  the  address.  (The term "octet" is used by Internet documentation for such 8-bit chunks.  The term "byte" is not used, because TCP/IP is supported  by  some computers that have byte sizes other than 8 bits.) Generally the structure of the  address  gives  you  some  information about  how  to  get  to  the  system.  For example, 128.6 is a network number assigned by a central authority to Rutgers University.  Rutgers uses  the  next  octet  to  indicate  which of the campus Ethernets is involved.  128.6.4 happens to be an  Ethernet  used  by  the  Computer Science  Department.    The last octet allows for up to 254 systems on each Ethernet.  (It is 254 because 0 and  255  are  not  allowed,  for reasons  that  will  be  discussed  later.)  Note that 128.6.4.194 and 128.6.5.194 would be different systems.  The structure of an  Internet address is described in a bit more detail later. Of  course  we  normally  refer  to  systems  by  name, rather than by Internet address.  When we specify a name, the network software  looks it  up  in  a  database,  and comes up with the corresponding Internet address.  Most of the network software deals strictly in terms of  the                                   6   address.  (RFC 882 describes the name server technology used to handle this lookup.) TCP/IP is  built  on  "connectionless"  technology.    Information  is transfered  as  a sequence of "datagrams".  A datagram is a collection of data that is sent as a single message.  Each of these datagrams  is sent  through  the network individually.  There are provisions to open connections (i.e.  to start a conversation that will continue for some time).    However at some level, information from those connections is broken up into datagrams, and  those  datagrams  are  treated  by  the network  as  completely  separate.    For example, suppose you want to transfer a 15000 octet file.  Most networks can't handle a 15000 octet datagram.   So the protocols will break this up into something like 30 500-octet datagrams.  Each of these datagrams  will  be  sent  to  the other  end.    At  that point, they will be put back together into the 15000-octet file.  However while those datagrams are in  transit,  the network doesn't know that there is any connection between them.  It is perfectly possible  that  datagram  14  will  actually  arrive  before datagram  13.    It is also possible that somewhere in the network, an error will occur, and some datagram won't get through at all.  In that case, that datagram has to be sent again. Note  by  the way that the terms "datagram" and "packet" often seem to be nearly interchangable.  Technically, datagram is the right word  to use  when  describing  TCP/IP.  A datagram is a unit of data, which is what the protocols deal with.  A packet is a physical thing, appearing on an Ethernet or some wire.  In most cases a packet simply contains a datagram, so there is  very  little  difference.    However  they  can differ.  When TCP/IP is used on top of X.25, the X.25 interface breaks the datagrams up into 128-byte packets.   This  is  invisible  to  IP, because  the  packets  are put back together into a single datagram at the other end before being processed by TCP/IP.  So in this case,  one IP  datagram  would  be carried by several packets.  However with most media, there are efficiency advantages to  sending  one  datagram  per packet, and so the distinction tends to vanish.   2.1 The TCP level  Two separate protocols are involved in handling TCP/IP datagrams.  TCP (the "transmission control protocol") is responsible for  breaking  up the  message  into  datagrams,  reassembling  them  at  the other end, resending anything that gets lost, and  putting  things  back  in  the right  order.  IP (the "internet protocol") is responsible for routing individual datagrams.  It may seem like TCP is  doing  all  the  work. And  in  small networks that is true.  However in the Internet, simply getting a datagram to its  destination  can  be  a  complex  job.    A connection  may require the datagram to go through several networks at Rutgers, a serial line to the John von Neuman Supercomputer Center,  a couple  of Ethernets there, a series of 56Kbaud phone lines to another NSFnet site, and more Ethernets on another campus.  Keeping  track  of the  routes  to all of the destinations and handling incompatibilities among different transport media turns out to be a complex job.    Note                                   7   that  the  interface  between TCP and IP is fairly simple.  TCP simply hands IP a datagram with a destination.   IP  doesn't  know  how  this datagram relates to any datagram before it or after it. It  may  have occurred to you that something is missing here.  We have talked about Internet addresses, but not about how you keep  track  of multiple  connections  to  a given system.  Clearly it isn't enough to get a datagram to the right  destination.    TCP  has  to  know  which connection  this  datagram  is  part  of.  This task is referred to as "demultiplexing."  In fact, there are several levels of demultiplexing going  on in TCP/IP.  The information needed to do this demultiplexing is contained in a series of "headers".  A header is simply a few extra octets  tacked  onto  the  beginning of a datagram by some protocol in order to keep track of it.  It's a lot like putting a letter  into  an envelope  and  putting  an  address  on  the  outside of the envelope. Except with modern networks it happens several times.  It's  like  you put the letter into a little envelope, your secretary puts that into a somewhat bigger envelope, the campus mail center  puts  that  envelope into a still bigger one, etc.  Here is an overview of the headers that get stuck on a message that passes through a typical TCP/IP network: We start with a single data stream, say a file you are trying to  send to some other computer:      ...................................................... TCP  breaks  it  up into manageable chunks.  (In order to do this, TCP has to know how large a datagram your network can handle.    Actually, the TCP's at each end say how big a datagram they can handle, and then they pick the smallest size.)      ....   ....   ....   ....   ....   ....   ....   .... TCP puts a header at the front of each datagram.  This header actually contains  at least 20 octets, but the most important ones are a source and destination "port number" and  a  "sequence  number".    The  port numbers  are used to keep track of different conversations.  Suppose 3 different people are transferring files.  Your TCP might allocate port numbers 1000, 1001, and 1002 to these transfers.  When you are sending a datagram, this becomes the "source" port number, since you  are  the source  of  the  datagram.    Of  course  the TCP at the other end has assigned a port number of its own for the conversation.  Your TCP  has to  know the port number used by the other end as well.  (It finds out when the connection starts, as we will explain below.)  It  puts  this in  the  "destination" port field.  Of course if the other end sends a datagram back to you, the source and destination port numbers will  be reversed,  since  then  it  will  be  the  source  and you will be the destination.  Each datagram has a sequence number.  This  is  used  so that  the  other  end  can make sure that it gets the datagrams in the right  order,  and  that  it  hasn't  missed  any.    (See   the   TCP specification for details.)  TCP doesn't number the datagrams, but the octets.  So if there are 500 octets of  data  in  each  datagram,  the first datagram might be numbered 0, the second 500, the next 1000, the next 1500, etc.  Finally, I will mention the  Checksum.    This  is  a number  that  is  computed by adding up all the octets in the datagram                                   8   (more or less - see the TCP spec).  The result is put in  the  header. TCP  at  the other end computes the checksum again.  If they disagree, then something bad happened to the datagram in transmission, and it is thrown away.  So here's what the datagram looks like now.     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+     |          Source Port          |       Destination Port        |     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+     |                        Sequence Number                        |     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+     |                    Acknowledgment Number                      |     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+     |  Data |           |U|A|P|R|S|F|                               |     | Offset| Reserved  |R|C|S|S|Y|I|            Window             |     |       |           |G|K|H|T|N|N|                               |     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+     |           Checksum            |         Urgent Pointer        |     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+     |   your data ... next 500 octets                               |     |   ......                                                      | If  we abbreviate the TCP header as "T", the whole file now looks like this:    T....   T....   T....   T....   T....   T....   T.... You will note that there are items in  the  header  that  I  have  not described  above.    They  are  generally  involved  with managing the connection.  In order to make sure the datagram  has  arrived  at  its destination,  the  recipient  has  to  send back an "acknowledgement". This is a datagram whose "Acknowledgement number" field is filled  in. For  example,  sending  a  packet  with  an  acknowledgement  of  1500 indicates that you have received all the data up to octet number 1500. If  the  sender  doesn't  get  an  acknowledgement within a reasonable amount of time, it sends the data  again.    The  window  is  used  to control  how  much  data can be in transit at any one time.  It is not practical to wait for each datagram to be acknowledged before  sending the  next  one.    That would slow things down too much.  On the other hand, you can't just keep sending, or a fast  computer  might  overrun the  capacity  of  a slow one to absorb data.  Thus each end indicates how much new data it is currently prepared to absorb  by  putting  the number  of  octets  in  its  "Window" field.  As the computer receives data, the amount of space left in its window decreases.  When it  goes to  zero, the sender has to stop.  As the receiver processes the data, it increases its window, indicating that it is ready  to  accept  more data.  Often the same datagram can be used to acknowledge receipt of a set of data and to give permission for  additional  new  data  (by  an updated  window).  The "Urgent" field allows one end to tell the other to skip ahead in its processing to a particular octet.  This is  often useful  for  handling asynchronous events, for example when you type a control character or other command that interrupts output.  The  other fields are beyond the scope of this document.                                     9   2.2 The IP level  TCP  sends each of these datagrams to IP.  Of course it has to tell IP the Internet address of the computer at the other end.  Note that this is  all  IP  is concerned about.  It doesn't care about what is in the datagram, or even in the TCP header.  IP's job is  simply  to  find  a route for the datagram and get it to the other end.  In order to allow gateways or other intermediate systems to  forward  the  datagram,  it adds  its  own  header.  The main things in this header are the source and destination Internet address (32-bit addresses, like 128.6.4.194), the  protocol  number,  and  another  checksum.    The source Internet address is simply the address of your machine.  (This is necessary  so the  other  end  knows where the datagram came from.)  The destination Internet address is the address  of  the  other  machine.    (This  is necessary  so  any  gateways  in  the  middle  know where you want the datagram to go.)  The protocol number tells IP at  the  other  end  to send  the  datagram  to TCP.  Although most IP traffic uses TCP, there are other protocols that can use IP, so you  have  to  tell  IP  which protocol  to send the datagram to.  Finally, the checksum allows IP at the other end to verify that the header  wasn't  damaged  in  transit. Note  that TCP and IP have separate checksums.  IP needs to be able to verify that the header didn't get damaged in transit, or it could send a  message to the wrong place.  For reasons not worth discussing here, it is both more efficient and safer to have  TCP  compute  a  separate checksum  for  the  TCP  header  and  data.  Once IP has tacked on its header, here's what the message looks like:     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+     |Version|  IHL  |Type of Service|          Total Length         |     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+     |         Identification        |Flags|      Fragment Offset    |     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+     |  Time to Live |    Protocol   |         Header Checksum       |     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+     |                       Source Address                          |     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+     |                    Destination Address                        |     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+     |  TCP header, then your data ......                            |     |                                                               | If we represent the IP header by an "I",  your  file  now  looks  like this:      IT....   IT....   IT....   IT....   IT....   IT....   IT.... Again,  the  header contains some additional fields that have not been discussed.  Most of them are beyond the scope of this document.    The flags  and fragment offset are used to keep track of the pieces when a datagram has to be split up.   This  can  happen  when  datagrams  are forwarded through a network for which they are too big.  (This will be discussed a bit more below.)  The time to live is  a  number  that  is decremented  whenever  the  datagram passes through a system.  When it goes to zero, the datagram is discarded.  This is done in case a  loop                                   10   develops  in the system somehow.  Of course this should be impossible, but  well-designed  networks  are  built  to  cope  with  "impossible" conditions. At this point, it's possible that no more headers are needed.  If your computer happens to have a direct phone  line  connecting  it  to  the destination  computer,  or  to  a  gateway,  it  may  simply  send the datagrams out on the line (though likely a synchronous  protocol  such as  HDLC  would be used, and it would add at least a few octets at the beginning and end).   2.3 The Ethernet level  However most of our networks these days use Ethernet.  So now we  have to  describe  Ethernet's headers.  Unfortunately, Ethernet has its own addresses.  The people who designed Ethernet wanted to make sure  that no  two  machines  would  end  up  with  the  same  Ethernet  address. Furthermore, they  didn't  want  the  user  to  have  to  worry  about assigning  addresses.    So  each  Ethernet  controller  comes with an address builtin from the factory.  In order to  make  sure  that  they would  never have to reuse addresses, the Ethernet designers allocated 48 bits for the Ethernet address.  People who make Ethernet  equipment have  to  register  with  a  central  authority, to make sure that the numbers they assign don't overlap any other manufacturer.  Ethernet is a "broadcast medium".  That is, it is in effect like an old party line telephone.  When you send a packet out on the Ethernet, every  machine on  the  network sees the packet.  So something is needed to make sure that the right machine gets it.  As you might guess, this involves the Ethernet  header.    Every  Ethernet packet has a 14-octet header that includes the source and destination Ethernet address, and a type code. Each machine is supposed to pay attention only to packets with its own Ethernet address in the destination field.  (It's  perfectly  possible to  cheat,  which  is  one reason that Ethernet communications are not terribly secure.)  Note  that  there  is  no  connection  between  the Ethernet address and the Internet address.  Each machine has to have a table of what Ethernet address corresponds to what  Internet  address. (We  will  describe  how  this  table is constructed a bit later.)  In addition to the addresses, the header contains a type code.  The  type code is to allow for several different protocol families to be used on the same network.  So you can use TCP/IP, DECnet, Xerox  NS,  etc.  at the  same  time.   Each of them will put a different value in the type field.  Finally,  there  is  a  checksum.    The  Ethernet  controller computes a checksum of the entire packet.  When the other end receives the packet, it recomputes the checksum, and throws the packet away  if the  answer  disagrees  with the original.  The checksum is put on the end of the packet, not in the header.  The final result is  that  your message looks like this:                                       11       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+     |       Ethernet destination address (first 32 bits)            |     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+     | Ethernet dest (last 16 bits)  |Ethernet source (first 16 bits)|     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+     |       Ethernet source address (last 32 bits)                  |     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+     |        Type code              |     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+     |  IP header, then TCP header, then your data                   |     |                                                               |         ...     |                                                               |     |   end of your data                                            |     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+     |                       Ethernet Checksum                       |     +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ If  we  represent  the  Ethernet  header  with  "E",  and the Ethernet checksum with "C", your file now looks like this:      EIT....C   EIT....C   EIT....C   EIT....C   EIT....C When these packets are received by the other end, of  course  all  the headers  are  removed.    The  Ethernet interface removes the Ethernet header and the checksum.  It looks at the type code.  Since  the  type code  is the one assigned to IP, the Ethernet device driver passes the datagram up to IP.  IP removes the IP header.   It  looks  at  the  IP protocol  field.    Since  the  protocol  type  is  TCP, it passes the datagram up to TCP.  TCP now looks at the sequence number.    It  uses the  sequence  numbers  and  other  information  to  combine  all  the datagrams into the original file. The ends our initial summary of TCP/IP.  There are still some  crucial concepts we haven't gotten to, so we'll now go back and add details in several areas.  (For detailed descriptions of the items discussed here see,  RFC  793  for  TCP,  RFC  791  for IP, and RFC's 894 and 826 for sending IP over Ethernet.)   3. Well-known sockets and the applications layer  So far, we have described how a stream  of  data  is  broken  up  into datagrams,  sent  to another computer, and put back together.  However something more is needed  in  order  to  accomplish  anything  useful. There  has  to  be  a  way for you to open a connection to a specified computer, log into it, tell it what file you  want,  and  control  the transmission  of  the  file.   (If you have a different application in mind, e.g. computer mail, some analogous protocol is needed.)  This is done  by  "application  protocols".  The application protocols run "on top" of TCP/IP.  That is, when they want to send a message, they  give the  message  to  TCP.   TCP makes sure it gets delivered to the other end.  Because TCP and IP take care of all the networking details,  the                                   12   applications  protocols can treat a network connection as if it were a simple byte stream, like a terminal or phone line. Before going into more details about applications programs, we have to describe how you find an application.  Suppose you want to send a file to a computer whose Internet address  is  128.6.4.7.    To  start  the process,  you  need  more than just the Internet address.  You have to connect to the FTP server at the  other  end.    In  general,  network programs  are  specialized  for a specific set of tasks.  Most systems have separate programs  to  handle  file  transfers,  remote  terminal logins, mail, etc.  When you connect to 128.6.4.7, you have to specify that you want to talk to the FTP server.    This  is  done  by  having "well-known  sockets"  for  each  server.    Recall that TCP uses port numbers to keep track of  individual  conversations.    User  programs normally  use more or less random port numbers.  However specific port numbers are assigned to the programs that sit  waiting  for  requests. For  example,  if  you  want  to send a file, you will start a program called "ftp".  It will open a connection using some random number, say 1234,  for  the  port number on its end.  However it will specify port number 21 for the other end.  This is the official port number for the FTP server.  Note that there are two different programs involved.  You run ftp on your side.  This is a program designed to  accept  commands from  your  terminal  and  pass them on to the other end.  The program that you talk to on the other machine  is  the  FTP  server.    It  is designed  to  accept commands from the network connection, rather than an interactive terminal.  There is no need for your program to  use  a well-known  socket  number  for  itself.  Nobody is trying to find it. However the servers have to have well-known numbers,  so  that  people can  open  connections  to  them and start sending them commands.  The official  port  numbers  for  each  program  are  given  in  "Assigned Numbers". Note  that  a  connection is actually described by a set of 4 numbers: the Internet address at each end, and the TCP port number at each end. Every  datagram  has  all  four of those numbers in it.  (The Internet addresses are in the IP header, and the TCP port numbers  are  in  the TCP header.)  In order to keep things straight, no two connections can have the same set of numbers.  However it is enough for any one number to  be  different.    For  example,  it  is perfectly possible for two different users on a machine to be sending files  to  the  same  other machine.    This  could  result  in  connections  with  the  following parameters:                    Internet addresses         TCP ports     connection 1  128.6.4.194, 128.6.4.7      1234, 21     connection 2  128.6.4.194, 128.6.4.7      1235, 21 Since the same machines are involved, the Internet addresses  are  the same.    Since  they  are  both  doing  file transfers, one end of the connection involves the well-known port number  for  FTP.    The  only thing  that  differs is the port number for the program that the users are running.  That's enough of a difference.  Generally, at least  one end  of  the  connection asks the network software to assign it a port number that is guaranteed to be unique.   Normally,  it's  the  user's end, since the server has to use a well-known number.                                   13   Now  that  we  know  how  to  open  connections, let's get back to the applications programs.  As mentioned earlier, once TCP  has  opened  a connection,  we  have  something  that might as well be a simple wire. All the hard parts are handled by TCP and IP.  However we  still  need some  agreement  as  to  what we send over this connection.  In effect this is simply an agreement on what set of  commands  the  application will  understand,  and  the  format  in  which  they  are  to be sent. Generally, what is sent is a combination of commands and data.    They use  context  to  differentiate.  For example, the mail protocol works like this: Your mail program opens a connection to the mail server  at the  other end.  Your program gives it your machine's name, the sender of the message, and the recipients you want it sent to.  It then sends a  command saying that it is starting the message.  At that point, the other end  stops  treating  what  it  sees  as  commands,  and  starts accepting  the  message.  Your end then starts sending the text of the message.  At the end of the message, a special mark is sent (a dot  in the first column).  After that, both ends understand that your program is again sending commands.  This is the simplest way to do things, and the one that most applications use. File  transfer  is  somewhat more complex.  The file transfer protocol involves two different connections.  It starts  out  just  like  mail. The user's program sends commands like "log me in as this user", "here is my password", "send me the file with this name".  However once  the command  to  send  data is sent, a second connection is opened for the data itself.  It would certainly be possible to send the data  on  the same  connection,  as  mail does.  However file transfers often take a long time.  The designers of the  file  transfer  protocol  wanted  to allow  the  user  to  continue  issuing commands while the transfer is going on.  For example, the user might make an inquiry,  or  he  might abort  the  transfer.    Thus  the designers felt it was best to use a separate connection for  the  data  and  leave  the  original  command connection  for  commands.    (It  is  also  possible  to open command connections to two different computers, and tell them to send  a  file from  one  to  the other.  In that case, the data couldn't go over the command connection.) Remote terminal connections use another mechanism still.   For  remote logins,  there  is just one connection.  It normally sends data.  When it is necessary to send a command (e.g. to set the terminal type or to change  some  mode),  a special character is used to indicate that the next character is a command.  If the user happens to type that special character as data, two of them are sent. We  are  not  going to describe the application protocols in detail in this document.  It's better to read the RFC's yourself.  However there are  a  couple of common conventions used by applications that will be described here.  First, the common network representation:  TCP/IP  is intended  to  be  usable  on  any  computer.    Unfortunately, not all computers agree on how data is represented.  There are differences  in character  codes  (ASCII  vs.  EBCDIC),  in  end  of  line conventions (carriage return, line feed, or a representation using counts), and in whether  terminals expect characters to be sent individually or a line at a time.   In  order  to  allow  computers  of  different  kinds  to communicate,   each   applications   protocol   defines   a   standard                                   14   representation.    Note  that  TCP  and  IP  do  not  care  about  the representation.    TCP  simply  sends octets.  However the programs at both ends have to agree on how the octets are to be interpreted.   The RFC  for  each  application  specifies the standard representation for that application.  Normally it  is  "net  ASCII".    This  uses  ASCII characters,  with end of line denoted by a carriage return followed by a line feed.  For remote login,  there  is  also  a  definition  of  a "standard terminal", which turns out to be a half-duplex terminal with echoing happening on the local machine.  Most applications  also  make provisions  for  the  two  computers to agree on other representations that they may find more convenient.  For example, PDP-10's have 36-bit words.    There  is a way that two PDP-10's can agree to send a 36-bit binary file.  Similarly, two systems that prefer full-duplex  terminal conversations  can  agree  on  that.    However each application has a standard representation, which every machine must support.   3.1 An example application: SMTP  In order to give a bit better idea what is involved in the application protocols,  I'm  going  to  show an example of SMTP, which is the mail protocol.  (SMTP is "simple mail transfer protocol.)  We assume that a computer called TOPAZ.RUTGERS.EDU wants to send the following message.   Date: Sat, 27 Jun 87 13:26:31 EDT   From: hedrick@topaz.rutgers.edu   To: levy@red.rutgers.edu   Subject: meeting   Let's get together Monday at 1pm. First,  note  that the format of the message itself is described by an Internet standard (RFC 822).  The standard specifies the fact that the message  must be transmitted as net ASCII (i.e. it must be ASCII, with carriage return/linefeed to delimit lines).   It  also  describes  the general  structure, as a group of header lines, then a blank line, and then the body of the message.  Finally, it describes the syntax of the header  lines in detail.  Generally they consist of a keyword and then a value. Note  that  the  addressee  is  indicated   as   LEVY@RED.RUTGERS.EDU. Initially,  addresses were simply "person at machine".  However recent standards have made things more flexible.  There  are  now  provisions for  systems  to handle other systems' mail.  This can allow automatic forwarding on behalf of computers not connected to the Internet.    It can be used to direct mail for a number of systems to one central mail server.  Indeed there is no requirement that an actual computer by the name  of RED.RUTGERS.EDU even exist.  The name servers could be set up so that you mail to department names, and each  department's  mail  is routed  automatically to an appropriate computer.  It is also possible that the part before the @ is something other than a user name.  It is possible  for  programs  to be set up to process mail.  There are also provisions  to  handle  mailing  lists,  and  generic  names  such  as                                   15   "postmaster" or "operator". The  way  the  message is to be sent to another system is described by RFC's 821 and 974.  The program that is going to be doing the  sending asks  the  name server several queries to determine where to route the message.  The first query is to find out which  machines  handle  mail for  the  name RED.RUTGERS.EDU.  In this case, the server replies that RED.RUTGERS.EDU handles its own mail.  The program then asks  for  the address of RED.RUTGERS.EDU, which is 128.6.4.2.  Then the mail program opens a TCP connection to port 25  on  128.6.4.2.    Port  25  is  the well-known  socket  used  for receiving mail.  Once this connection is established, the mail program starts sending  commands.    Here  is  a typical  conversation.  Each line is labelled as to whether it is from TOPAZ or RED.  Note that TOPAZ initiated the connection:     RED    220 RED.RUTGERS.EDU SMTP Service at 29 Jun 87 05:17:18 EDT     TOPAZ  HELO topaz.rutgers.edu     RED    250 RED.RUTGERS.EDU - Hello, TOPAZ.RUTGERS.EDU     TOPAZ  MAIL From:<hedrick@topaz.rutgers.edu>     RED    250 MAIL accepted     TOPAZ  RCPT To:<levy@red.rutgers.edu>     RED    250 Recipient accepted     TOPAZ  DATA     RED    354 Start mail input; end with <CRLF>.<CRLF>     TOPAZ  Date: Sat, 27 Jun 87 13:26:31 EDT     TOPAZ  From: hedrick@topaz.rutgers.edu     TOPAZ  To: levy@red.rutgers.edu     TOPAZ  Subject: meeting     TOPAZ     TOPAZ  Let's get together Monday at 1pm.     TOPAZ  .     RED    250 OK     TOPAZ  QUIT     RED    221 RED.RUTGERS.EDU Service closing transmission channel First, note that commands all use normal text.  This is typical of the Internet  standards.    Many  of  the  protocols  use  standard  ASCII commands.  This makes it easy  to  watch  what  is  going  on  and  to diagnose  problems.  For example, the mail program keeps a log of each conversation.  If something goes wrong, the log  file  can  simply  be mailed  to  the  postmaster.  Since it is normal text, he can see what was going on.  It also allows a human to interact  directly  with  the mail  server,  for  testing.  (Some newer protocols are complex enough that this is not practical.  The commands would have to have a  syntax that would require a significant parser.  Thus there is a tendency for newer protocols to use binary formats.  Generally they are  structured like  C or Pascal record structures.)  Second, note that the responses all begin with numbers.  This is also typical of  Internet  protocols. The  allowable  responses  are  defined  in the protocol.  The numbers allow the user program to respond unambiguously.    The  rest  of  the response  is  text,  which is normally for use by any human who may be watching or looking at a log.  It has no effect on  the  operation  of the  programs.  (However there is one point at which the protocol uses part of the text of the response.)   The  commands  themselves  simply allow  the  mail  program  on  one  end  to  tell  the mail server the                                   16   information it needs to know in order to deliver the message.  In this case,  the  mail  server  could  get the information by looking at the message itself.  But for more complex cases, that would not  be  safe. Every  session  must  begin  with  a HELO, which gives the name of the system that initiated the connection.  Then the sender and  recipients are specified.  (There can be more than one RCPT command, if there are several recipients.)  Finally the data itself is sent.  Note that  the text  of the message is terminated by a line containing just a period. (If such a line appears in the message, the period is doubled.)  After the  message  is  accepted,  the  sender  can send another message, or terminate the session as in the example above. Generally, there is a pattern to the response numbers.   The  protocol defines  the  specific set of responses that can be sent as answers to any given command.  However programs that don't want to  analyze  them in  detail  can  just  look at the first digit.  In general, responses that begin with a 2  indicate  success.    Those  that  begin  with  3 indicate  that some further action is needed, as shown above.  4 and 5 indicate errors.  4 is a "temporary" error, such as  a  disk  filling. The  message should be saved, and tried again later.  5 is a permanent error, such as a  non-existent  recipient.    The  message  should  be returned to the sender with an error message. (For  more  details about the protocols mentioned in this section, see RFC's 821/822 for mail, RFC 959 for file transfer, and  RFC's  854/855 for  remote  logins.  For the well-known port numbers, see the current edition of Assigned Numbers, and possibly RFC 814.)   4. Protocols other than TCP: UDP and ICMP  So far, we have described only connections that use TCP.  Recall  that TCP  is  responsible  for  breaking  up  messages  into datagrams, and reassembling them properly.  However in  many  applications,  we  have messages  that  will  always  fit in a single datagram.  An example is name lookup.  When a user attempts to make  a  connection  to  another system,  he  will  generally  specify  the system by name, rather than Internet address.  His system has to translate that name to an address before  it  can  do  anything.  Generally, only a few systems have the database used to translate names to addresses.  So the  user's  system will want to send a query to one of the systems that has the database. This query is going to be very short.  It will certainly  fit  in  one datagram.    So  will the answer.  Thus it seems silly to use TCP.  Of course TCP does more than just break things up  into  datagrams.    It also  makes  sure  that  the  data  arrives, resending datagrams where necessary.  But for a question that fits  in  a  single  datagram,  we don't  need  all the complexity of TCP to do this.  If we don't get an answer after a few seconds, we can just ask again.   For  applications like this, there are alternatives to TCP. The most common alternative is UDP ("user datagram protocol").  UDP is designed for applications where you don't need  to  put  sequences  of datagrams  together.  It fits into the system much like TCP.  There is                                   17   a UDP header.  The network software puts the UDP header on  the  front of  your  data, just as it would put a TCP header on the front of your data.  Then UDP sends the data  to  IP,  which  adds  the  IP  header, putting  UDP's  protocol number in the protocol field instead of TCP's protocol number.  However UDP doesn't do as much  as  TCP  does.    It doesn't  split data into multiple datagrams.  It doesn't keep track of what it has sent so it can resend if necessary.  About  all  that  UDP provides  is  port  numbers,  so  that several programs can use UDP at once.  UDP port numbers are used just like TCP port  numbers.    There are  well-known  port numbers for servers that use UDP.  Note that the UDP header is shorter than a TCP header.   It  still  has  source  and destination  port  numbers,  and  a checksum, but that's about it.  No sequence number, since it is not needed.  UDP is used by the protocols that  handle  name  lookups (see IEN 116, RFC 882, and RFC 883), and a number of similar protocols. Another  alternative  protocol  is  ICMP  ("Internet  control  message protocol").    ICMP  is  used  for  error messages, and other messages intended for the TCP/IP software itself, rather  than  any  particular user  program.  For example, if you attempt to connect to a host, your system may get back an ICMP message saying "host unreachable".    ICMP can  also be used to find out some information about the network.  See RFC 792 for details of ICMP.  ICMP is  similar  to  UDP,  in  that  it handles messages that fit in one datagram.  However it is even simpler than UDP.  It doesn't even have port numbers in its header.  Since all ICMP  messages are interpreted by the network software itself, no port numbers are needed to say where a ICMP message is supposed to go.   5. Keeping track of names and information: the domain system  As we indicated earlier, the network software generally needs a 32-bit Internet  address  in  order  to open a connection or send a datagram. However users prefer to deal with computer names rather than  numbers. Thus  there  is  a database that allows the software to look up a name and find the corresponding number.  When the Internet was small,  this was  easy.  Each system would have a file that listed all of the other systems, giving both their name and number.  There are  now  too  many computers  for  this  approach to be practical.  Thus these files have been replaced by a set of name servers that keep track of  host  names and  the corresponding Internet addresses.  (In fact these servers are somewhat more general than that.  This is just one kind of information stored in the domain system.)  Note that a set of interlocking servers are used, rather than a single central one.  There  are  now  so  many different  institutions  connected  to  the  Internet that it would be impractical for them to  notify  a  central  authority  whenever  they installed  or moved a computer.  Thus naming authority is delegated to individual institutions.  The name servers form a tree,  corresponding to  institutional  structure.    The names themselves follow a similar structure.  A typical example is the name BORAX.LCS.MIT.EDU.  This  is a  computer  at  the Laboratory for Computer Science (LCS) at MIT.  In order to find its Internet address,  you  might  potentially  have  to consult  4  different  servers.  First, you would ask a central server                                   18   (called the root) where the EDU server is.  EDU is a server that keeps track of educational institutions.  The root server would give you the names and Internet addresses of several servers for EDU.   (There  are several  servers  at  each  level,  to allow for the possibly that one might be down.)  You would then ask EDU where the server for  MIT  is. Again,  it  would  give  you  names  and Internet addresses of several servers for MIT.  Generally, not all of those servers would be at MIT, to  allow for the possibility of a general power failure at MIT.  Then you would ask MIT where the server for LCS is, and finally  you  would ask one of the LCS servers about BORAX.  The final result would be the Internet address for BORAX.LCS.MIT.EDU.    Each  of  these  levels  is referred  to  as  a  "domain".  The entire name, BORAX.LCS.MIT.EDU, is called a "domain name".    (So  are  the  names  of  the  higher-level domains, such as LCS.MIT.EDU, MIT.EDU, and EDU.) Fortunately,  you  don't really have to go through all of this most of the time.  First of all, the root name servers also happen to  be  the name  servers  for  the  top-level domains such as EDU.  Thus a single query to a root  server  will  get  you  to  MIT.    Second,  software generally  remembers answers that it got before.  So once we look up a name at LCS.MIT.EDU, our software remembers where to find servers  for LCS.MIT.EDU,  MIT.EDU,  and EDU.  It also remembers the translation of BORAX.LCS.MIT.EDU.  Each of these pieces of information has a "time to live"  associated with it.  Typically this is a few days.  After that, the information expires and has to be looked up again.    This  allows institutions to change things. The  domain  system  is not limited to finding out Internet addresses. Each domain name is a node in a database.  The node can  have  records that  define  a number of different properties.  Examples are Internet address, computer type, and a list of services provided by a computer. A  program  can  ask  for  a  specific  piece  of  information, or all information about a given name.  It is possible  for  a  node  in  the database  to  be  marked as an "alias" (or nickname) for another node. It is also possible to use the  domain  system  to  store  information about users, mailing lists, or other objects. There  is  an  Internet  standard  defining  the  operation  of  these databases, as well as the protocols used  to  make  queries  of  them. Every  network utility has to be able to make such queries, since this is now the official way to evaluate host names.   Generally  utilities will talk to a server on their own system.  This server will take care of contacting the other servers for them.  This keeps down the  amount of code that has to be in each application program. The  domain  system  is  particularly  important for handling computer mail.  There are entry types to define what computer handles mail  for a  given  name, to specify where an individual is to receive mail, and to define mailing lists. (See RFC's 882, 883, and 973 for specifications of the domain  system. RFC 974 defines the use of the domain system in sending mail.)                                     19   6. Routing  The   description  above  indicated  that  the  IP  implementation  is responsible for getting datagrams to the destination indicated by  the destination address, but little was said about how this would be done. The task of finding how to  get  a  datagram  to  its  destination  is referred to as "routing".  In fact many of the details depend upon the particular implementation.  However some general things can be said. First, it is necessary to understand the model on which IP  is  based. IP assumes that a system is attached to some local network.  We assume that the system can send datagrams to any  other  system  on  its  own network.    (In  the  case  of  Ethernet, it simply finds the Ethernet address of the destination system, and puts the datagram  out  on  the Ethernet.)    The  problem  comes  when  a  system  is asked to send a datagram to a system on a different network.  This problem is  handled by  gateways.   A gateway is a system that connects a network with one or more other networks.  Gateways  are  often  normal  computers  that happen  to have more than one network interface.  For example, we have a Unix machine that has two different Ethernet interfaces.  Thus it is connected  to networks 128.6.4 and 128.6.3.  This machine can act as a gateway between those two networks.  The software on that machine must be  set  up  so that it will forward datagrams from one network to the other.  That is, if a machine on network 128.6.4 sends a  datagram  to the  gateway,  and  the  datagram is addressed to a machine on network 128.6.3, the gateway will forward the  datagram  to  the  destination. Major communications centers often have gateways that connect a number of different  networks.    (In  many  cases,  special-purpose  gateway systems provide better performance or reliability than general-purpose systems acting as gateways.  A number of vendors sell such systems.) Routing in IP is  based  entirely  upon  the  network  number  of  the destination  address.    Each computer has a table of network numbers. For each network number, a gateway is listed.  This is the gateway  to be used to get to that network.  Note that the gateway doesn't have to connect directly to the network.  It just has to be the best place  to go  to  get there.  For example at Rutgers, our interface to NSFnet is at the John von Neuman Supercomputer Center (JvNC). Our connection  to JvNC  is  via  a  high-speed  serial line connected to a gateway whose address is 128.6.3.12.  Systems on net 128.6.3 will list 128.6.3.12 as the  gateway  for  many  off-campus  networks.  However systems on net 128.6.4 will list 128.6.4.1 as the gateway to  those  same  off-campus networks.    128.6.4.1  is  the  gateway  between networks 128.6.4 and 128.6.3, so it is the first step in getting to JvNC. When a computer wants to send a datagram, it first checks  to  see  if the  destination address is on the system's own local network.  If so, the datagram can be sent directly.  Otherwise, the system  expects  to find an entry for the network that the destination address is on.  The datagram is sent to the gateway listed in that entry.  This table  can get quite big.  For example, the Internet now includes several hundred individual networks.  Thus various strategies have been  developed  to reduce  the size of the routing table.  One strategy is to depend upon "default routes".  Often, there is only one gateway out of a  network.                                   20   This  gateway might connect a local Ethernet to a campus-wide backbone network.  In that case, we don't need to have  a  separate  entry  for every  network  in  the  world.    We  simply define that gateway as a "default".  When no specific  route  is  found  for  a  datagram,  the datagram  is  sent to the default gateway.  A default gateway can even be used when there are several gateways  on  a  network.    There  are provisions  for  gateways  to  send a message saying "I'm not the best gateway -- use this one instead."  (The message is sent via ICMP.  See RFC  792.)  Most network software is designed to use these messages to add entries to their routing tables.  Suppose network 128.6.4 has  two gateways, 128.6.4.59 and 128.6.4.1.  128.6.4.59 leads to several other internal Rutgers networks.  128.6.4.1 leads indirectly to the  NSFnet. Suppose  we  set  128.6.4.59  as  a default gateway, and have no other routing table entries.  Now what  happens  when  we  need  to  send  a datagram  to  MIT?    MIT  is  network 18.  Since we have no entry for network 18, the datagram will be sent to the default, 128.6.4.59.   As it  happens,  this  gateway  is the wrong one.  So it will forward the datagram to 128.6.4.1.  But it will also send back an error saying  in effect: "to get to network 18, use 128.6.4.1".  Our software will then add an entry to the routing table.  Any future datagrams to  MIT  will then  go  directly to 128.6.4.1.  (The error message is sent using the ICMP protocol.  The message type is called "ICMP redirect.") Most IP experts recommend that individual computers should not try  to keep  track  of  the  entire network.  Instead, they should start with default gateways, and let the gateways tell them the routes,  as  just described.   However this doesn't say how the gateways should find out about the routes.  The gateways can't depend upon this strategy.  They have  to  have fairly complete routing tables.  For this, some sort of routing protocol is needed.  A routing protocol is simply a  technique for  the  gateways  to  find each other, and keep up to date about the best way to get to every network.   RFC  1009  contains  a  review  of gateway  design  and  routing.    However rip.doc is probably a better introduction to the subject.  It contains some tutorial material,  and a detailed description of the most commonly-used routing protocol.   7. Details about Internet addresses: subnets and broadcasting  As  indicated earlier, Internet addresses are 32-bit numbers, normally written as 4 octets (in decimal), e.g. 128.6.4.7.  There are  actually 3  different types of address.  The problem is that the address has to indicate both the network and the host within the  network.    It  was felt  that  eventually  there would be lots of networks.  Many of them would be small, but probably 24 bits would be needed to represent  all the  IP  networks.  It was also felt that some very big networks might need 24 bits to represent all of their hosts.  This would seem to lead to  48  bit  addresses.  But the designers really wanted to use 32 bit addresses.  So they adopted a kludge.  The assumption is that most  of the  networks will be small.  So they set up three different ranges of address.  Addresses beginning with 1 to 126 use only the  first  octet for  the network number.  The other three octets are available for the host number.  Thus 24 bits are available for hosts.  These numbers are                                   21   used  for large networks.  But there can only be 126 of these very big networks.  The Arpanet is one, and there are a  few  large  commercial networks.    But  few  normal organizations get one of these "class A" addresses.  For normal large organizations, "class  B"  addresses  are used.    Class  B  addresses  use the first two octets for the network number.  Thus network numbers are 128.1 through 191.254.  (We avoid  0 and  255,  for  reasons  that  we  see below.  We also avoid addresses beginning with 127, because that is used by some systems  for  special purposes.)    The  last  two  octets  are available for host addesses, giving 16 bits of host address.   This  allows  for  64516  computers, which should be enough for most organizations.  (It is possible to get more than one class B address, if you run  out.)    Finally,  class  C addresses  use  three  octets,  in  the  range 192.1.1 to 223.254.254. These allow only 254 hosts on each network, but there can be  lots  of these  networks.   Addresses above 223 are reserved for future use, as class D and E (which are currently not defined). Many large organizations find it convenient to  divide  their  network number into "subnets".  For example, Rutgers has been assigned a class B address, 128.6.  We find it convenient to use the third octet of the address to indicate which Ethernet a host is on.  This division has no significance outside of Rutgers.  A computer  at  another  institution would treat all datagrams addressed to 128.6 the same way.  They would not look at the third octet of the address.   Thus  computers  outside Rutgers  would  not have different routes for 128.6.4 or 128.6.5.  But inside Rutgers, we treat 128.6.4 and 128.6.5 as separate networks.  In effect, gateways inside Rutgers have separate entries for each Rutgers subnet, whereas gateways outside  Rutgers  just  have  one  entry  for 128.6.  Note  that  we  could  do  exactly  the  same thing by using a separate class C address for each Ethernet.   As  far  as  Rutgers  is concerned,  it  would be just as convenient for us to have a number of class C addresses.  However using class C addresses would make  things inconvenient for the rest of the world.  Every institution that wanted to talk to us would have to have a separate entry for each one of  our networks.   If every institution did this, there would be far too many networks for any reasonable gateway to keep track of.  By  subdividing a  class B network, we hide our internal structure from everyone else, and  save  them  trouble.    This  subnet  strategy  requires  special provisions in the network software.  It is described in RFC 950. 0  and  255  have  special  meanings.  0 is reserved for machines that don't know their address.  In certain circumstances it is possible for a  machine not to know the number of the network it is on, or even its own host address.  For example, 0.0.0.23 would be a machine that  knew it was host number 23, but didn't know on what network. 255  is  used for "broadcast".  A broadcast is a message that you want every system on the network to see.    Broadcasts  are  used  in  some situations  where you don't know who to talk to.  For example, suppose you need to look  up  a  host  name  and  get  its  Internet  address. Sometimes  you  don't know the address of the nearest name server.  In that case, you might send the request as a broadcast.  There are  also cases  where a number of systems are interested in information.  It is then less expensive to send a single broadcast than to send  datagrams individually  to  each host that is interested in the information.  In                                   22   order to send a broadcast, you use an address that is  made  by  using your  network  address, with all ones in the part of the address where the host number goes.  For example, if you are on network 128.6.4, you would   use   128.6.4.255  for  broadcasts.    How  this  is  actually implemented depends upon the medium.   It  is  not  possible  to  send broadcasts  on the Arpanet, or on point to point lines.  However it is possible on an Ethernet.  If you use an Ethernet address with all  its bits  on (all ones), every machine on the Ethernet is supposed to look at that datagram. Although the official broadcast address for  network  128.6.4  is  now 128.6.4.255,  there  are  some  other addresses that may be treated as broadcasts by certain implementations.  For convenience, the  standard also  allows  255.255.255.255 to be used.  This refers to all hosts on the local network.  It is often simpler to use 255.255.255.255 instead of  finding out the network number for the local network and forming a broadcast address such as 128.6.4.255.   In  addition,  certain  older implementations  may  use  0  instead  of  255  to  form the broadcast address.    Such  implementations  would  use  128.6.4.0  instead   of 128.6.4.255  as  the  broadcast  address on network 128.6.4.  Finally, certain older implementations may not understand about subnets.   Thus they consider the network number to be 128.6.  In that case, they will assume a broadcast address  of  128.6.255.255  or  128.6.0.0.    Until support  for  broadcasts is implemented properly, it can be a somewhat dangerous feature to use. Because 0 and 255 are used for unknown and broadcast addresses, normal hosts  should never be given addresses containing 0 or 255.  Addresses should never begin with 0, 127, or any number above  223.    Addresses violating these rules are sometimes referred to as "Martians", because of rumors that the Central University of Mars is using network 225.   8. Datagram fragmentation and reassembly  TCP/IP is designed for use  with  many  different  kinds  of  network. Unfortunately,  network  designers  do not agree about how big packets can be.  Ethernet packets can be 1500 octets long.    Arpanet  packets have  a  maximum  of around 1000 octets.  Some very fast networks have much larger packet sizes.  At first, you might think  that  IP  should simply  settle  on  the  smallest  possible size.  Unfortunately, this would cause serious performance problems.    When  transferring  large files, big packets are far more efficient than small ones.  So we want to be able to use the largest packet size possible.  But we also  want to  be  able  to  handle  networks  with  small limits.  There are two provisions for this.  First, TCP has the ability to "negotiate"  about datagram  size.  When a TCP connection first opens, both ends can send the maximum datagram size they can  handle.    The  smaller  of  these numbers  is  used  for  the  rest  of the connection.  This allows two implementations that can handle big datagrams to use  them,  but  also lets  them  talk  to  implementations that can't handle them.  However this doesn't completely solve the problem.  The most  serious  problem is  that the two ends don't necessarily know about all of the steps in                                   23   between.  For example, when sending data between Rutgers and Berkeley, it is likely that both computers will be on Ethernets.  Thus they will both  be  prepared  to  handle  1500-octet  datagrams.    However  the connection will at some point end up going over the Arpanet.  It can't handle packets of that size.  For this reason, there are provisions to split   datagrams   up   into   pieces.    (This  is  referred  to  as "fragmentation".)  The IP header  contains  fields  indicating  the  a datagram  has  been split, and enough information to let the pieces be put back together.  If a gateway connects an Ethernet to the  Arpanet, it must be prepared to take 1500-octet Ethernet packets and split them into pieces that will fit on the Arpanet.    Furthermore,  every  host implementation  of  TCP/IP  must  be prepared to accept pieces and put them back together.  This is referred to as "reassembly". TCP/IP implementations differ in the approach they take to deciding on datagram  size.    It  is  fairly  common  for  implementations to use 576-byte datagrams whenever they can't verify that the entire path  is able  to  handle larger packets.  This rather conservative strategy is used because of the number of implementations with bugs in the code to reassemble  fragments.    Implementors  often try to avoid ever having fragmentation occur.  Different implementors take different approaches to  deciding  when  it  is safe to use large datagrams.  Some use them only for the local network.  Others will use them for any  network  on the   same   campus.    576  bytes  is  a  "safe"  size,  which  every implementation must support.   9. Ethernet encapsulation: ARP  There was a brief discussion earlier about what IP datagrams look like on  an  Ethernet.    The  discussion  showed  the  Ethernet header and checksum.  However it left one hole: It didn't say how to  figure  out what Ethernet address to use when you want to talk to a given Internet address.  In fact, there is a separate protocol for this,  called  ARP ("address  resolution protocol").  (Note by the way that ARP is not an IP protocol.  That is, the ARP datagrams  do  not  have  IP  headers.) Suppose  you  are  on  system  128.6.4.194  and you want to connect to system 128.6.4.7.  Your system will first verify that 128.6.4.7 is  on the  same network, so it can talk directly via Ethernet.  Then it will look up 128.6.4.7 in its ARP table, to see if  it  already  knows  the Ethernet  address.    If  so, it will stick on an Ethernet header, and send the packet.  But suppose this system is not  in  the  ARP  table. There  is  no  way  to  send the packet, because you need the Ethernet address.  So it  uses  the  ARP  protocol  to  send  an  ARP  request. Essentially  an  ARP  request  says  "I  need the Ethernet address for 128.6.4.7".  Every system listens to ARP requests.  When a system sees an  ARP  request  for itself, it is required to respond.  So 128.6.4.7 will see the request, and will respond with an  ARP  reply  saying  in effect "128.6.4.7 is 8:0:20:1:56:34".  (Recall that Ethernet addresses are 48 bits.  This is 6 octets.  Ethernet addresses are conventionally shown  in  hex,  using  the punctuation shown.)  Your system will save this information in its ARP table, so future packets will go directly. Most  systems  treat the ARP table as a cache, and clear entries in it                                   24   if they have not been used in a certain period of time. Note by the way that ARP requests must be sent as "broadcasts".  There is  no  way  that  an  ARP  request  can be sent directly to the right system.  After all, the whole reason for sending  an  ARP  request  is that  you  don't know the Ethernet address.  So an Ethernet address of all ones is  used,  i.e.  ff:ff:ff:ff:ff:ff.    By  convention,  every machine  on  the Ethernet is required to pay attention to packets with this as an address.  So every system sees every ARP  requests.    They all  look to see whether the request is for their own address.  If so, they respond.  If not, they could just ignore it.   (Some  hosts  will use  ARP  requests  to update their knowledge about other hosts on the network, even if the request isn't for them.)  Note that packets whose IP  address  indicates broadcast (e.g. 255.255.255.255 or 128.6.4.255) are also sent with an Ethernet address that is all ones.   10. Getting more information  This directory contains  documents  describing  the  major  protocols. There  are literally hundreds of documents, so we have chosen the ones that seem most important.  Internet standards are called RFC's.    RFC stands  for  Request  for  Comment.   A proposed standard is initially issued as a proposal, and given an RFC number.   When  it  is  finally accepted,  it is added to Official Internet Protocols, but it is still referred to by the RFC number.   We  have  also  included  two  IEN's. (IEN's  used  to  be  a  separate  classification  for  more  informal documents.  This classification no longer exists -- RFC's are now used for  all  official  Internet documents, and a mailing list is used for more informal reports.)  The convention is that  whenever  an  RFC  is revised, the revised version gets a new number.  This is fine for most purposes, but it causes problems with two documents: Assigned  Numbers and  Official  Internet  Protocols.  These documents are being revised all the time, so the RFC number keeps changing.  You will have to look in rfc-index.txt to find the number of the latest edition.  Anyone who is seriously interested in TCP/IP should read the  RFC  describing  IP (791).    RFC 1009 is also useful.  It is a specification for gateways to be used by NSFnet.  As such, it contains an overview of  a  lot  of the  TCP/IP technology.  You should probably also read the description of at least one of the application protocols, just to get a  feel  for the  way  things  work.    Mail is probably a good one (821/822).  TCP (793) is of course a very basic specification.  However  the  spec  is fairly  complex,  so  you should only read this when you have the time and patience to think about it carefully.  Fortunately, the author  of the  major  RFC's  (Jon Postel) is a very good writer.  The TCP RFC is far easier to read than you would expect, given the complexity of what it  is  describing.    You  can  look at the other RFC's as you become curious about their subject matter. Here is a list of the documents you are more likely to want:      rfc-index list of all RFC's                                   25        rfc1012   somewhat fuller list of all RFC's      rfc1011   Official Protocols.  It's useful to scan  this  to  see                what tasks protocols have been built for.  This defines                which  RFC's  are  actual  standards,  as  opposed   to                requests for comments.      rfc1010   Assigned  Numbers.  If you are working with TCP/IP, you                will probably want a hardcopy of this as  a  reference.                It's  not  very  exciting  to  read.   It lists all the                offically defined well-known ports and  lots  of  other                things.      rfc1009   NSFnet  gateway  specifications.  A good overview of IP                routing and gateway technology.      rfc1001/2 netBIOS: networking for PC's      rfc973    update on domains      rfc959    FTP (file transfer)      rfc950    subnets      rfc937    POP2: protocol for reading mail on PC's      rfc894    how IP is to be put on Ethernet, see also rfc825      rfc882/3  domains (the database used to go  from  host  names  to                Internet  address  and back -- also used to handle UUCP                these days).  See also rfc973      rfc854/5  telnet - protocol for remote logins      rfc826    ARP - protocol for finding out Ethernet addresses      rfc821/2  mail      rfc814    names and ports - general  concepts  behind  well-known                ports      rfc793    TCP      rfc792    ICMP      rfc791    IP      rfc768    UDP      rip.doc   details of the most commonly-used routing protocol      ien-116   old  name  server  (still  needed  by  several kinds of                system)      ien-48    the  Catenet  model,   general   description   of   the                                   26                  philosophy behind TCP/IP  The following documents are somewhat more specialized.      rfc813    window and acknowledgement strategies in TCP      rfc815    datagram reassembly techniques      rfc816    fault isolation and resolution techniques      rfc817    modularity and efficiency in implementation      rfc879    the maximum segment size option in TCP      rfc896    congestion control      rfc827,888,904,975,985                EGP and related issues  To those of you who may be reading this document remotely  instead  of at  Rutgers:  The  most  important  RFC's  have  been collected into a three-volume set, the DDN Protocol Handbook.  It is available from the DDN  Network  Information  Center,  SRI  International, 333 Ravenswood Avenue, Menlo Park, California 94025 (telephone: 800-235-3155).    You should  be able to get them via anonymous FTP from sri-nic.arpa.  File names are:     RFC's:     rfc:rfc-index.txt     rfc:rfcxxx.txt   IEN's:     ien:ien-index.txt     ien:ien-xxx.txt rip.doc is available  by  anonymous  FTP  from  topaz.rutgers.edu,  as /pub/tcp-ip-docs/rip.doc. Sites with access to UUCP but not FTP may be able to retreive them via UUCP from UUCP host rutgers.  The file names would be    RFC's:     /topaz/pub/pub/tcp-ip-docs/rfc-index.txt     /topaz/pub/pub/tcp-ip-docs/rfcxxx.txt   IEN's:     /topaz/pub/pub/tcp-ip-docs/ien-index.txt     /topaz/pub/pub/tcp-ip-docs/ien-xxx.txt   /topaz/pub/pub/tcp-ip-docs/rip.doc Note that SRI-NIC has the entire set of RFC's and IEN's,  but  rutgers and topaz have only those specifically mentioned above.                                       27