Fall, 2001 CS 640 1

Network Applications

OutlineFile Transfer ProtocolTelnetSimple Mail Transfer ProtocolPeer-to-peer applications (Napster,Gnutella)

Fall, 2001 CS 640 2

Overview

• So far we have introduced general notions of networking infrastructure, organization and functionality

• Now we will begin our task of understanding networks at a deeper level by focusing on applications– Software which facilitate interaction between

humans and the network

• Today’s lecture will focus on simple apps and peer-to-peer apps

Fall, 2001 CS 640 3

File Transfer Protocol• Most basic file transfer application in the Internet

– One of the original client/server applications run on the ARPANET• Runs on top of TCP• Allows for both file transfer and interactive access• Requires authentication via user name and password• Requires that a host system run an FTP server

– Listens for incoming requests on a well known port (21)– Anonymous/Guest logins are common

• FTP is a two process model– Control process which communicates with peer control process

• These processes communicate commands/responses as well as port information

– Data transfer process which actually transfers requested file

Fall, 2001 CS 640 4

File Transfer Protocol contd.• Client control process connects to server control

process– ftp mirror1.cs.wisc.edu

• The client also starts a data transfer process which listens on a local port– Communicates this port number to server via control

process

• If client requests a file transfer, server initiates connection to client’s data transfer port– Server uses well known port for data transfer (20)

• Commands used by FTP are actually a subset of TELNET protocol NVT ASCII

Fall, 2001 CS 640 5

FTP exampletux29(1)% ftp mirror1.cs.wisc.eduConnected to mirror1.cs.wisc.edu.220 Welcome to The UW-Madison Computer Sciences mirror site530 Please login with USER and PASS.Name (mirror1.cs.wisc.edu:jgast): anonymous331 Please specify the password.Password: XXXXXXX230 Login successful. Have fun.Remote system type is UNIX.Using binary mode to transfer files.ftp> cd pub250 Directory successfully changed.ftp> ls200 PORT command successful. Consider using PASV.150 Here comes the directory listing.drwxrwxrwx 2 0 0 2048 Aug 23 12:56 mirrors226 Directory send OK.ftp> quit221 Goodbye.tux29(2)%

3-digitcode easesparsing

5xx meansactionrequired

1xx=OK, I will2xx=OK, done.3xx=OK so far4xx=NO, temp

Fall, 2001 CS 640 6

Typical FTP Session

listen on port 21 Cntrl

211025Connect to port 21

Data

Cntrl

Cntrl Cntrl

Data

DataData

CLIENT SERVER

Fall, 2001 CS 640 7

Typical FTP Session

Cntrl

201493 Data

Cntrl

Cntrl Cntrl

Data

DataData

1025 21Ask for LIST, 1493

List of files

PORT command OK

1025 21RETR gnotella.zip, 1522

PORT command OK

201522

Lots of data

CLIENT SERVER

Never talks to an unpriviledged port on an untrusted server

Fall, 2001 CS 640 8

Passive FTP Session

Cntrl

211025

Data

Cntrl

Cntrl Cntrl

Data

DataData

CLIENT SERVERPASV

Passive Mode, 4107

41071673

ask toLIST files

211025

Use passive mode if the server can’t initiate a connection to the client

The list of files

fire

wall

Fall, 2001 CS 640 9

Telnet• Internet’s most basic network virtual terminal application

allowing users to log into a remote host• Runs on top of TCP• Requires authentication via user name and password• Requires host system to run a telnet server (telnetd)• Passes keystrokes to remote system and carries output back

to user’s screen• Offers three basic services

– Network virtual terminal – standard interface to remote system– Mechanism for negotiating communication options– Treats both ends of a connection symmetrically

Fall, 2001 CS 640 10

Telnet• Client’s TELNET process connects to server’s TELNET

– telnet foo.cs.wisc.edu• Server listens on well known port (23) for incoming

connections – forks new slave process to handle connection

• Much of the hard work in telnet is to accommodate heterogeneity of systems– Control characters, etc.– Accomplished via network virtual terminal (NVT) specification for

ASCII• It’s not very secure!• It can cause a lot of traffic - tinygrams

– Nagle’s algorithm (RFC 896, ‘84)• Collects and sends groups of data based on ACK process

Fall, 2001 CS 640 11

Simple Mail Transfer Protocol• Basic protocol for email exchange over the Internet• Runs on top of TCP• Fundamental difference between SMTP and FTP/TELNET is that

it is NOT an interactive protocol– Messages are queued and spooled by SMTP agent

• Users interact with email application– Lots!

• Application interfaces with Message Transfer Agent– Sendmail on UNIX– Setup and configured by admins.

• SMTP specifies how MTA’s pass email across the Internet– Also uses NVT commands

Fall, 2001 CS 640 12

Simple Mail Transfer Protocol• Client uses email application to construct and send messages

– Addresses consist of machine name and mailbox address• Mailbox is usually the same as one’s login but you can have aliases• Destination machine can also be an alias@cs.wisc.edu

• Message is passed to mail spooler which is part of MTA– Application communicates with MTA via email transfer protocol

• Post Office Protocol (POP3) is common, but not very secure• Our department uses IMAP

• MTA’s on remote systems listen for incoming mail on well known port (25)

• Messages are delivered in two parts – header and body– Header format has exact specification = RFC 822– Body content types are specified by MIME

Fall, 2001 CS 640 13

SMTP Example

Jim GastUW-madison

Anne GastHumko Products

From:jgast@cs.wisc.eduTo: "Anne Gast" <AGast@humko.com>Subject: I hope you brought a toothbrushDate: Tues, 11 Sep 2001 10:46:16 -0400MIME-Version: 1.0X-Mailer: Internet Mail Service (5.5.1960.3)Content-Type: text/plainContent-Transfer-Encoding: 7bit

Hi, Anne!

You won’t be flying home today. Hijackedplanes hit both towers of the World TradeCenter!

-Jim

header

body

Fall, 2001 CS 640 14

Sample SMTP Sessiontux34(21)% telnet smtp.cs.wisc.edu 25Trying 128.105.6.11...Connected to schroeder.cs.wisc.edu (128.105.6.11).220 schroeder.cs.wisc.edu ESMTP Sendmail 8.11.3; Tue, 11 Sep 2001 14:09:52 -0500HELO jgast.cs.wisc.edu250 schroeder.cs.wisc.edu Hello tux34.cs.wisc.edu [128.105.111.134], pleased to meet youMAIL FROM:<jgast@cs.wisc.edu>250 2.1.0 <jgast@cs.wisc.edu> ... Sender okRCPT TO:<jgast@tds.net>250 2.1.5 <jgast@tds.net>... Recipient okDATA354 Enter mail, end with "." on a line by itselfTo: jgast@tds.net

Test for CS640.250 2.0.0. f8BJAeq14849 Message accepted for deliveryQUIT221 2.0.0 schroeder.cs.wisc.edu closing connectionConnection closed by foreign host.tux34(22)%

Envelope

Header

Body

Fall, 2001 CS 640 15

Peer to Peer Applications• Applications which facilitate the sharing of data between widely

distributed users– Clearly the new “killer app” on the Internet– Napster model

• One central source of meta data – Gnutella model

• Distributed source of meta data• This is the way things will go…

• Napster– Invented by Shawn Fanning at Northeastern U. in 1998-9– Founded as a company in May, 1999– Indexes millions of songs on millions of systems world wide

• Well known servers at Abovenet and globalcenter– Sued by Recording Industry Association of America in Dec. 1999– Currently in limbo…

Fall, 2001 CS 640 16

Napster Protocol• There are three basic entities in Napster

– Client – browses, requests files– Server – end host system which shares/serves files– Metaserver/Redirector – file index repository

• Protocol is “private” – has been reverse engineered• Messages are passed between entities

– Form: <length><type><data>• Length and type are two bytes• Length specifies the number of bytes of data• Data is ASCII string often enclosed in quotes

– There are a large number of possible messages defined for the protocol

• Information – errors, versions, status…• Control – login, upgrades, add/delete accounts…• Operation – search, listings, transfer requests, etc.

Fall, 2001 CS 640 17

Napster Protocol contd.• “Official” software runs on Windows

– Many other clients available eg. Gnapster– All you need to access Napster is a client

• Runs on TCP– Servers (typically) use ports 7777, 8888– Metaservers run on port 8875

• Protocol– Connect to a well known metaserver and login– Can join chat session or search for files (our focus)– Clients send “search query” request - results in locations of

matching files– Downloading files happen directly between hosts

• Metaserver is not involved

Fall, 2001 CS 640 18

Napster Protocol contd.• File transfers use one of four modes

– Up/download– Firewall up/download

• Non-firewalled download– Client sends download (203) request to metaserver– Metaserver responds with download ack (204) that contains

user, IP address, port … of requested file– Client makes TCP connection to server

• Server responds with ASCII “1” immediately– Client sends GET in one packet then username, filename

• Also include byte offset to resume interrupted transfer

Fall, 2001 CS 640 19

Napster Protocol contd.• Remote client returns filesize followed by datastream• Client notifies metaserver transfer has begun (218) and ended

(219)• Firewalled download

– 204 ack from metaserver has port number 0 indicating server is behind a firewall

– Client sends firewall download request (500) to metaserver requesting that server upload file to client

• Client then waits for server to upload requested file– Metaserver then contacts server requesting upload to client

• Done by HTTP tunneling: send message which looks like HTTP request– Upload continues just like download

• Each upload is tracked

Fall, 2001 CS 640 20

Gnutella• Napster’s flaw is that there is a single source for the index• The simple idea in Gnutella is to distribute the method for

finding data– Great idea!– Lots of fun architectural possibilities!– A few interesting research projects underway

• Gnutella is a distributed search protocol with a decentralized model– Clients can issue/view query results– Clients can serve/request data– Clients accept queries and respond with matches from their local

data store– Provides for highly reliable system

Fall, 2001 CS 640 21

Gnutella Protocol• Protocol defines method of client communication

– Set of descriptors used for communicating data– Setof rules governing inter-client exchange of descriptors

• Descriptors– Ping: active discovery of hosts on a network– Pong: response to Ping includes client address and metadata– Query: search mechanism– QueryHit: response to Query includes info necessary to get data– Push: mechanism enabling firewalled clients to be part of network

• A Gnutella client (servent) connects to network by establishing a connection with another client on the network– Finding another client is not part of Gnutella spec.

• Host cache services are the typical way this is done

Fall, 2001 CS 640 22

Gnutella Protocol• New client then creates connection to the Gnutella client and

thereby becomes part of the network– Gnutella client can reject the connect request– Successful new client can then send/receive descriptors

• Pings/pongs are then sent to establish network– No specification as to how much/often to probe– Network data can/is cached

• Message routing should be well behaved– Ping/Query descriptors should be sent to all directly connected

clients– Pong/QueryHit descriptors should be sent back along same path– TTL is mechanism to limit distance

• File downloads via HTTP/1.0 protocol via direct connect

Fall, 2001 CS 640 23

Other features• http://www.KaZaA.com

– “Distributed, self-organizing”– Only supernodes are search hubs– Intelligent download (Any of several identical items)– Simultaneous download of chunks

• http://www.napigator.com– Finds and organizes open napster servers

• http://www.filetopia.org– Adds encryption

• http://www.audiogalaxy.com• http://www.cubicmetercrystal.com/alpine