Home Map Index Search News Archives Links About LF
[Top bar]
[Bottom bar]
This article is available in: English  Castellano  

convert to palmConvert to GutenPalm
or to PalmDoc

[Photo of the Author]
by Guido Socher

About the author:

Guido is a long time Linux fan. All his computers in the house are integrated into a network.


Content:

Home Networking, glossary and overview

[Illustration]

Abstract:

Networking becomes more and more popular. Your old 486 can perfectly work as file-server or run IP-Masquerading to give the whole family Internet access from different computers at the same time. This article will provide you with the basic overview to start a home network. It will explain the vocabulary which you need to understand literature and other articles on networking. The focus of the article is a small scale home network.



 

10Base2

This is an Ethernet with 10MBit/s over coaxial cable. The ends of the cable must be terminated on both ends with a 50 Ohm resistor. The total length of the cable may not be more than 185 meters. The minimum distance between the cards should be more than 0.5 meters. The network cards are connected to the cable using a T-connector. No Hub is required and the cards are all connected directly to the cable.This is probably the cheapest setup for a home network. It is also called "cheapernet". It is important that the cable runs directly from network card to network card (bus topology). There must not be any cable between the card and the T-connector.
10Base2
 

10baseT

Ethernet with 10MBit/s using twisted pair cable. Every card is connected via twisted pair cable to a central point called hub. This results in a star topology of the network. The cable length between card and hub should not be more than 100 meters. If you want to connect only 2 cards then it is also possible to leave away the hub and use a cross-connect cable.
10BaseT  

100baseT and 100baseT4

This type of network uses also twisted pair cable (as 10baseT ) but runs with 100 MBit/s. For most home applications the cheaper 10MBit/s networks should be more than sufficient. 100baseT uses pairs 2 and 3 from the twisted pair cable. 100baseT4 uses all 4 wire pairs.  

RJ-45 Connector

This is the type of connector used for the 10baseT, 100baseT and 100baseT4 type of networks. It is a small flat plastic connector with 8 pins.
RJ-45
Pin Name Description for 10baseT
1 TX+ Tranceive Data+
2 TX- Tranceive Data-
3 RX+ Receive Data+
4 n/c 100BaseT4 only
5 n/c 100BaseT4 only
6 RX- Receive Data-
7 n/c 100BaseT4 only
8 n/c 100BaseT4 only
  Note: TX & RX are
swapped on Hub's.

Note: you need special pliers to connect the RJ-45 connector to the cable.

 

Cable types

The following cable types are common: The different Twisted Pair pair cable types are standardized in categories according to their attenuation.

The mapping between the pins on the RJ-45 connector and the Twisted Pair wires is standardized. There are several standards and they use slightly different colors. Valid pin assignments are:
Wire pair number Connector pins standard 1 standard 2 standard 3 standard 4
1 4/5 blue/white white/blue white/blue white/brown
2 3/6 white/green red/orange cyan/purple green/yellow
3 1/2 white/orange black/green white/orange grey/pink
4 7/8 white/brown yellow/brown cyan/purple blue/red

Cabling is really only relevant if you want to run 20 or more meters of cable through the whole house. If you want to connect just a few computers in the same room then you can get cheap ready made cables with connectors. These are typically available with a length between 1 and 10 meters.

 

Cross connect cable

This is a special cable to connect just 2 network cards over twisted pair cable. You need only a HUB if you want to connect 3 and more Computers. The cross connect swaps RX and TX. The 2 connectors are connected as follows:
Pin No Pin No
1 connected to 3
2 connected to 6
3 connected to 1
4 connected to 5 (100baseT4)
5 connected to 4 (100baseT4)
6 connected to 2
7 connected to 8 (100baseT4)
8 connected to 7 (100baseT4)
9 connected to 9 (100baseT4)
You can also buy ready made cross connect cables with connectors.  

Hub

A HUB is needed for networks with twisted pair cable and more than 2 computers. The HUB amplifies the signal from a computer and distributes it to all other connected computers. For home networking you typically want a 5 or 8 port "work group hub". Ethernet switches can also be used instead of HUBs but they are far more expensive and make only sense if you expect high traffic with many computers connected to the network.  

LAN

LAN stands for Local Area Network and denotes a broadcast network. A broadcast network is a network where each computer can reach each other computer without going through a router. E.g all computers connected along one coaxial cable or all computers connected to one HUB form a LAN.  

Cards

My personal experience is that the cheapest cards are the best. The more expensive ones do usually not give you any additional features. ISA PNP cards may cause trouble. Most cards which are sold as ISA PNP cards come with a DOS floppy disk and software where you can change the flash ROM settings of that card and configure it to not use PNP. You can set a fixed interrupt and IO address. (Note: if you have a PNP bios then you must also tell it to disable PNP for this interrupt). The standard IO address is normally io=0x300. You can see which address and interrupts are already used by typing:
cat /proc/interrupts /proc/ioports | more
(Note: These are the interrupts and addresses which are really used by a driver at the moment.)

If you bought such an ISA PNP card then you will probably find that the DOS floppy disk is not of much use for a Linux only user. You or a friend might have an old boot-able dos floppy somewhere. Copy the program that came with the network card to that floppy. Boot from it and change the setting of the network card. The settings are stored in flash Rom so, you just do this once after you installed the hardware and then never again.

The cheapest cards are usually NE2000 cards and cost today about $15 (or 15 Euro). These are well supported under Linux. NE2000 cards are available as ISA or PCI bus cards.
The ISA bus cards need the kernel modules "ne" and "8390". You can either let the driver probe for it or provide IRQ and IO at the command line. Loading of kernel module without auto-probing can be done as follows:
insmod 8390
insmod ne irq=10 io=0x300

With a PCI card you don't usually have any trouble with interrupts and IO addresses but they are a bit more expensive. The kernel modules for the NE2000 PCI cards are called ne2k-pci and 8390.

 

Network interface

The network interface is physically represented by your network card but the term network interface is also used for the representation of a software name to which you assign an IP-address. If you have a modem and an Ethernet card on your computer then you have already at least two IP addresses and network interfaces for this one computer. An IP addresses is always assigned to a network interface never to a host (computer). You can uniquely drive the hostname from the IP address but there is in general a one to many relation between hostname and IP address. The command ifconfig is used to print the settings for the different active network interfaces and you can use this command also to configure a network interface. Here is an example printout:
> /sbin/ifconfig -a
lo        Link encap:Local Loopback
          inet addr:127.0.0.1  Bcast:127.255.255.255  Mask:255.0.0.0
          UP BROADCAST LOOPBACK RUNNING  MTU:3584  Metric:1
          RX packets:1664 errors:0 dropped:0 overruns:0
          TX packets:1664 errors:0 dropped:0 overruns:0

eth0      Link encap:10Mbps Ethernet  HWaddr 00:80:AD:71:52:19
          inet addr:10.0.0.1  Bcast:10.255.255.255  Mask:255.0.0.0
          UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
          RX packets:5993 errors:0 dropped:0 overruns:0
          TX packets:4073 errors:0 dropped:0 overruns:0
          Interrupt:5 Base address:0x300

dummy0    Link encap:10Mbps Ethernet  HWaddr 00:00:00:00:00:00
          inet addr:10.0.0.1  Bcast:10.255.255.255  Mask:255.0.0.0
          UP BROADCAST RUNNING NOARP MULTICAST  MTU:1500  Metric:1
          RX packets:0 errors:0 dropped:0 overruns:0
          TX packets:0 errors:0 dropped:0 overruns:0

ppp0      Link encap:Point-Point Protocol
          inet addr:199.94.253.30  P-t-P:199.94.253.4  Mask:255.255.255.0
          UP POINTOPOINT RUNNING  MTU:552  Metric:1
          RX packets:7 errors:0 dropped:0 overruns:0
          TX packets:8 errors:0 dropped:0 overruns:0
As you can see there are 4 network interfaces in the above example.  

Routing

Your Linux kernel has an "internal IP router". The routing table tells it what to do with an incoming IP packet. Dependent on the IP address of that packet you can decide in the routing table where to send it to. Under Linux you can type /sbin/route or netstat -r to see the routing table but the standard Unix command is netstat -r. How to set up the routing would be beyond the scope this article and there will be a separate article covering this subject. The Linux Network Andministators's Guide (www.linuxdoc.org/LDP/nag/nag.html) is one good online source for more information on how to set up routing.  

PLIP

Parallel port IP is a possibility to use the parallel port as network interface. The kernel configuration parameter is called CONFIG_PLIP and you should preferably configure it as module. PLIP is a cheap possibility to connect your Laptap to another computer and via this computer to your home network. You need to buy a parallel port cable with 2 male connectors also known as LapLink cable. If you want to build the cable your self then the wiring is:

Pin No Pin No
17 to 17
25 to 25
2 to 15
3 to 13
4 to 12
5 to 10
Pin No Pin No
6 to 11
15 to 2
13 to 3
12 to 4
10 to 5
11 to 6
PLIP is not as fast as Ethernet but the speed is already sufficient for decent networking applications like e.g NFS.

 

IP addresses

You need to decide which addresses to use for your network. Normally you will get one IP addresses assigned form your Internet service provider when you "go online". This IP addresses is more or less automatically assigned to the ppp interface. You do normally not have to worry about it. What you need to decide now is what addresses to assign to the other interfaces of your private network (e.g eth0 and plip1 interface). A number of addresses are reserved for private use. This addresses are not routed in the Internet and can not cause you any trouble when you re-use them. Their ranges are:
Netmask Network Addresses
255.0.0.0 10.0.0.0 - 10.255.255.255
255.255.0.0 172.16.0.0 - 172.31.255.255
255.255.255.0 192.168.0.0 - 192.168.255.255
The first block is a single Class A network number, the second block is a set of 16 contiguous Class B network numbers, and the third block is a set of 255 contiguous Class C network numbers.

Which address should you pick? Well, it does not really matter but I recommend using consecutive numbers. E.g you have 2 computers connected via Ethernet and you need now 2 addresses to assign to the 2 network cards (you assign it to the software representation of the cards. This software representation is called interface). Just use 192.168.0.1 and 192.168.0.2

 

Files

Most distributions come these days with pretty graphical user interfaces (GUIs) to configure networking (e.g yast or netcfg). You can use them if you want. Refer to the user manual of your Linux distribution on how to use them. This article shall not focus on a specific distribution. It explains therefore the underlying configuration files that will somehow magically be modified if you use the GUIs from your distribution. This gives you also the possibility to verify what the configuration GUIs did.

/etc/hosts and /etc/host.conf

For a small home network it does not make much sense to have DNS server running. Instead you can just enter the names of the machines in a file called /etc/hosts Example:

# ipaddr. fullname alias
127.0.0.1 localhost
192.168.0.1 obelix.mynet obelix
192.168.0.2 asterix.mynet asterix
192.168.0.3 idefix.mynet idefix
Together with the file /etc/hosts goes a file called /etc/host.conf where you say how to resolve names. You should use the following in this file:
# /etc/host.conf
order hosts bind
# Allow multiple addrs
multi on

/etc/resolv.conf

The file /etc/resolv.conf says where to resolve anything that can not be found in /etc/hosts. You need to ask your ISP for the ip address of his DNS server. You should enter more than one DNS server if possible. If the first DNS server is slow or down then the next will be asked. The syntax is:

#/etc/resolv.conf
#nameserver ipaddr.
nameserver 123.456.789.1
# alternative DNS server:
nameserver 123.456.111.9

setting the hostname

/etc/HOSTNAME or (/etc/sysconfig/network for redhat) is a file where many distributions configure the name of the local machine. But in reality the host name of a machine is set via the command hostname. What happens in these distributions is that during startup (e.g /etc/rc.d/rc.sysinit) the file /etc/HOSTNAME is read and then its content is feed to the command hostname. Which file is exactly used depends really on the peculiarities of your specific installation. It should however now be easy for you to investigate this since you know that you should grep for hostname.
Example how to set the hostname:

hostname obelix.mynet

tcp wrapper

/etc/hosts.allow this is a configuration file for a security system called tcp wrapper. All you need to know, is that you should add the ip-addresses of all the machines in your local network, excluding the machine where this file is located, behind the keyword ALL:.
Example: obelix has 2 network interfaces that go to other hosts in your local net. The interface with the ip address 192.168.0.2 goes to asterix and the interface with the ip address 192.168.0.3 goes to idefix. In this case you write:

# hosts.allow
# See tcpd(8) and hosts_access(5) for a description
ALL: 192.168.0.2   192.168.0.3
If you run a NFS server on obelix then you need also to add the line
portmap: 192.168.0.2   192.168.0.3
Network File System (NFS) gives you the possibility to access files on other host as if they were locally on the disk.

/etc/hosts.deny is also for the tcp wrapper it should never be changed and should always look as follows:

portmap: ALL
ALL: ALL

/etc/hosts.equiv

/etc/hosts.equiv allows local users to log into other machines (e.g with rsh) without password. You may decide if you want this. The syntax is: On full name (as in /etc/hosts) of a trusted host per line. E.g

asterix.mynet
idefix.mynet

/etc/hosts.lpd

The /etc/hosts.lpd file is important if you would like to enable network printing services via lpd. How to configure network printing is another story but this file must be present on the machine where you run the lp-daemon and it must list the names of all the machines that are allowed to print. The syntax is the same as hosts.equiv. The network printer configuration it self is best done with the system admin tools from your distribution (e.g yast in suse and printtool in redhat). The Printing-HOWTO from the Linux Documenation project describes how to set it up if you do not want to use the system admin tools from your distribution.

sendmail

The file /etc/service.switch is used by sendmail. On the hosts inside your network (the ones that do not have a direct Internet connection) you write:

hosts   files
aliases files
You configure then your sendmail on these hosts to forward all mail to the machine that has the modem and you use this one as a mailhost.

Sendmail reads at startup a configuration file called /etc/sendmail.cf and this file is quite difficult to read. The sendmail sources use therefore a macro language called m4 to generate the sendmail.cf out of a macro file. To do this you need the m4 macro processor (probably already installed on your system) and a part of the sendmail sources. Redhat Linux installs also these sources to /usr/lib/sendmail-cf other distributions may not come with it but you can always unpack the sendmail sources. Documentation on how the sendmail configuration works can be obtained from www.sendmail.org.

A central mail host is called mail-hub. The idea is that all the different machines in your network forward just their mail to this hub and the hub stores the mail. You then access it via NFS or pop/imap. A m4 file to do exactly that looks as follows:

#divert(-1)
include(`../m4/cf.m4')
VERSIONID(`mc file for asterix, A HOST INSIDE THE PRIVATE NETWORK')
undefine(`UUCP_RELAY')
undefine(`BITNET_RELAY')
OSTYPE(`linux')
# put here your outside mail domain. E.g if your user id is
# joe and your mail address [email protected] then you write:
MASQUERADE_AS(linuxnet.somewhere.org)
# Do not show these internal hostnames to the outside world:
MASQUERADE_DOMAIN(asterix.mynet)
MASQUERADE_DOMAIN(asterix)
MASQUERADE_DOMAIN(localdomain)
MASQUERADE_DOMAIN(localhost)
FEATURE(masquerade_envelope)
#the /etc/passwd entry: mail:*:8:12:mail:/var/spool/mail:
define(`confDEF_USER_ID',``8:12'')
# the centail mail hub is obelix, Note dot at the end
define(`MAIL_HUB', `obelix.mynet.')dnl
define(`confHOSTS_FILE',/etc/hosts)dnl
MAILER(smtp)dnl
MAILER(local)dnl
To generate the sendmail.cf file form the above sendmail.mc file you go into the cf directory of the sources (redhat /usr/lib/sendmail-cf/cf) and put the sendmail.mc there. Then you run m4 sendmail.mc > /etc/sendmail.cf

Please note: to just get networking running you will not need sendmail at all and you can send and receive mail with netscape without bothering about sendmail. Sendmail is probably something for the more advances Linux users.

A bit more difficult is the configuration of your mailhost. It must deliver local mail and forward external mail. External mail can e.g be forwarded via smtp (Simple Mail Transfer Protocol) in which case you should read the sendmail documentation ( www.sendmail.org) about "define(`SMART_HOST', smtp:provider.mail.host)dnl". For outgoing mail via smtp you need also on your mailhost computer a /etc/service.switch that enables DNS lookup:

hosts   files dns
aliases files
You can also use UUCP and other methods. It really depends on what your Internet provider offers.
UUCP stands for Unix to Unix Copy and is an old message transfer system but it is very good for getting mail from your ISP. Again your Internet provider must offer this service The more Windows focused companies may not even know what UUCP is. How does UUCP work for you? Example: The domain of your provider is somewhere.org and your host is called form outside linuxnet (outside means ppp-interface, internally in our examples we called it sofar obelix). Your provider gives you all mail via UUCP that goes to <sombody>@linuxnet.somewhere.org You can then distribute it using sendmail. Every local user has his/her own mail account which YOU can set up. It is just a matter of adding another unix user. Unfortunately most Internet providers offer today just one pop mail address and then you all this does not apply to you. Use in this case netscape or fetchmail.

 

IP-Masquerading (NAT)

IP Masquerading is the Linux term and most other people call it Network Address Translation. This is a very valuable thing for a small home network. It gives you the possibility to access the Internet from any of your local machines although only one machine is known in the Internet. This one machine is probably connected via modem/PPP and has a valid Internet wide routed ip-address assigned dynamically (or statically) to its ppp0 interface. With IP-Masquerading several people on different computers in your home network can access the Internet at the same time. LinuxFocus will have a separate article on how to setup IP-Masquerading. I recommend also reading the howto on IP-Masquerade. For the moment I will just explain how it works in principle.

To understand how IP-Masquerade works you must understand how a TCP connection is set up. When you connect to a host on the Internet with your web-browser then you connect to the ip address of that machine and the web server application. The application (e.g a web server) is selected by something called port number. The port numbers are pre-defined. Port 80 is the port number for the web server application. Thus if you type www.linuxfocus.org in your netscape then it knows that it must connect to the IP address 195.53.25.18 (= www.linuxfocus.org) and port 80.
For a full tcp connection where data can be send and received, pairs of IP address and port numbers are needed. That is 195.53.25.18 and 80 plus your own ip address and a local port number to which the web server can reply. The local port number is usually allocated at random. E.g:

www at www.linuxfocus.org:    your.machine.at.your.isp:
IP=195.53.25.18 Port=80 ----- IP=123.123.456.1 Port=4123
If you start a second netscape window and setup an other connection to www.linuxfocus.org then a new local port number is choosen:
www at www.linuxfocus.org:    your.machine.at.your.isp:
IP=195.53.25.18 Port=80 ----- IP=123.123.456.1 Port=4123
IP=195.53.25.18 Port=80 ----- IP=123.123.456.1 Port=4541
Note: twice port 80 is not correct but
we just say so to make the example simpler.
When a TCP packet comes back from the web server then it knows to which netscape window to go because of the difference in the port number on your machine. In the case of a private network behind your.machine.at.your.isp the situation is:
www at www.linuxfocus.org --- your.machine.at.your.isp --- privatehost1.mynet
                                                        |
                                                        +- privatehost2.mynet
None of the IP addresses of interfaces from privatehost1.mynet are known in the Internet. Therefore it is impossible to send an IP-packet from the Internet back to privatehost1.mynet. No router in the Internet would know where to send it to. However your.machine.at.your.isp has an IP addresses which is known in the Internet and as we saw above, different connections can be told apart by their port-numbers.
privatehost1.mynet can request a connection to www.linuxfocus.org and port 80. It would choose a port number, e.g 3211, and send a TCP packet in the direction of your.machine.at.your.isp. The host your.machine.at.your.isp would normally (without IP-Masquerade) just route the packet out to the Internet where it would disappear (at least the answer from the www.linuxfocus.org server would disappear). With IP-Masquerade your.machine.at.your.isp catches the packet and builds up a new connection with a local random port number. It then remembers that any answer packets back to that port must be sent on to privatehost1.mynet and port 3211.
So IP-Masquerade, builds up a new connection for you and has a mapping table from its port numbers to IP addresses and port numbers in the private network. If an answer packet comes back to the machine where IP-Masquerade is running it checks to which private IP addresses and portnumber it should forward the packet.

 

IRQ

Problems with interrupts are not specific to networking. They can occur anytime you add new hardware to your computer. Unfortunately the PC hardware does only allow for 16 interrupts and most of them are already in use. It is possible to share interrupts between two PCI cards at the expense of some CPU power. It is not possible to share interrupts between ISA bus cards. If you have a PNP Bios and insert a ISA bus card with fixed IRQ into your computer then you must first go into the BIOS setup and reserve that IRQ otherwise the BIOS may try to assign the IRQ to other hardware and that will not work.

Under Linux you can see how often certain hardware has already generated an interrupt by using the command cat /proc/interrupts. Hardware which is not used will not generate any interrupts.

The following is a list of the pre-defined IRQs. You can only use them for other purposes if you do not use the hardware component assigned to it. E.g if you use SCSI drives instead of IDE then you can use IRQ 14/15 for other things.
IRQ Assigned to
0 timer
1 keyboard
2 cascade (programmable IRQ controller)
3 serial 2
4 serial 1
5 free, usually sound card
6 floppy
7 parallel port, but under linux printers will not generate any iterrrupt
8 real time clock
9 free
10 free
11 free
12 ps2 mouse (/dev/psaux)
13 math co-processor
14 IDE 1
15 IDE 2

 

Example network

After so much theory I would like to discuss a little example network with you. Why do people build home networks? Here some ideas:

  1. Networks allow us to share resources. One computer my have a modem and Internet connection, all share it and use it. One computer may have a printer, all computers can share it. One computer may have a special drive (e.g zip-dirve, cd-rom, dvd) all computers can use it and mount it via NFS. One computer may have a large disk with lots of images and sound files. All computers can access it.
  2. It is a technically interesting thing.
Now to our example network. We have one computer with modem, one with a printer. These two are connected via Ethernet. Both have cdrom and floppy drives. There is also a laptop which has neither printer nor modem and also no cdrom. We connect it to the free parallel port on the first computer and give it therefore access to printer cdrom and the modem/Intenet.
example network How would you set up such a network? Well, first you buy the hardware. You need [besides computers modem and printer], 2 network cards a cross-connect cable and a Laplink cable. No hub is required. You install the hardware connect the cables and make sure you cause no interrupt (IRQ) conflict with the new hardware. Next you either re-configure the kernel or you just load the necessary kernel modules if they are already there. Most hardware support will be already compiled in if you use the pre-configured kernel from one of the major distributions. Now you use the command ifconfig and assign an IP address to the new network interfaces. You then use the command route to define which IP-packets must be send where. After that your network should be operational. You can not yet use domain names and IP Maskerade will also not yet work but you can send around a ping and watch how the LED on your network card flashes.
If you use the following IP addresses:
IP 
192.168.0.1 eth0 on Workstation 1
192.168.0.2 eth0 on Workstation 2
Then you can type on Workstation 1 ping 192.168.0.2:
>ping 192.168.0.2
PING 192.168.0.2 (192.168.0.2): 56 data bytes
64 bytes from 192.168.0.2: icmp_seq=0 ttl=64 time=0.9 ms
64 bytes from 192.168.0.2: icmp_seq=1 ttl=64 time=0.8 ms
64 bytes from 192.168.0.2: icmp_seq=2 ttl=64 time=0.9 ms

--- ping statistics ---
3 packets transmitted, 3 packets received, 0% packet loss
round-trip min/avg/max = 0.8/0.8/0.9 ms
You stop this with crtl-c. Take also a look at the command cat /proc/interrupts where you should see your network card or PLIP in the case of the Laptop connection. A cat /proc/ioports should show your new hardware and the command netstat -i shows how many IP packets were sent to which interface.
You can use at this stage already ftp between the computers and transfer files.

Basically your network is there but nothing really cool works. Now you must edit the different configuration files (use also the nice graphical tools that are available with the various distributions). Finally you setup IP-Maskerade and NFS. Now your network is fully functional.

I hope that this gave you a good overview. Don't be disappointed if I did not tell you exactly how to really configure the software part. There will be separate articles covering ifconfig/route/netstat, IP-Maskerade, PPP setup, network printing with lpd and NFS. This article was supposed to give a good overview to make it easier to understand other articles on networking.


Happy networking

ps: LinuxFocus is always happy to receive articles on networking. Not all articles are written yet and this is just the start of a series. Let us know if you are an experienced user and would like to write an article.

 

References

 

Talkback form for this article

Every article has its own talkback page. On this page you can submit a comment or look at comments from other readers:
 talkback page 

Webpages maintained by the LinuxFocus Editor team
© Guido Socher, FDL
LinuxFocus.org

Click here to report a fault or send a comment to LinuxFocus

2001-01-27, generated by lfparser version 2.8