Promoting Linux Requires Advertising. It Matters to Me.
Linux Network Address Translation
Network Address Translation (NAT) is a vitally important
Internet technology for a variety of reasons. It can
provide load balancing for parallel processing, it can
provide several types of strong access security, and it can
provide fault-tolerance and high-availability. Finally,
it can simplify some basic network administration functions.
Below, we sketch the possible uses, and then follow up with
- Internet Traffic Load Balancing
- IBM describes a router which they used for parallelizing web
server queries for the Olympic Games web server. If I understand
the trade press properly, this router takes a TCP/IP connection
request, and re-labels it, redistributes it to one of many
web servers ("mirrors") operating at different IP addresses.
Each server maintains an identical set of web pages. The user
is unaware of the existence of multiple web servers/mirrors, as
they (i.e. their browser) connect to the externally published,
well-known domain name. The mirrors were geographically distributed
(Atlanta, New York, California), and requests were routed to the
least-busy and/or ping-time closest server. Although the above
description is technically vague, the utility of such a technology
- Intranet Compute-Server Load Balancing
- The web is not the only client-server technology that can
potentially have trouble when there are too many clients
trying to access the same server. As an example, a database
server may be trying to fulfill database queries. Since
database queries are much, much more CPU intensive than simple
web-page queries, it is easier to overload a database.
If, however, the database queries are all "read-only" queries
(they do not modify the database), then it is possible to
distribute the load to several machines. Network Address Translation
(NAT) can provide the mechanism. With NAT, the headers of IP
packets that come into one machine are re-written, and forwarded
to the least-busy database server in the cluster.
The reply packets from the servers are again re-written and
returned to the client, thus making it appear that there
was only one database server with only one IP address.
Note that such a scheme provides not only load-balancing and
improved performance, but it also provides fault-tolerance:
individual servers can be taken off-line and serviced, while
the overall system continues to operate without stopping.
- Firewall Security through Masquerading
- One important security concept is that it is much easier to
guard a single point of entry than it is to guard many points.
This is the principle behind the Internet firewall: a single
machine that divides the network into the "inside" and the
"outside", with all traffic passing through the firewall.
By protecting the single network firewall, the entire
internal network can be protected. Masquerading allows
insiders to get out, without allowing outsiders in.
Masquerading re-writes the IP headers of internal packets
going out, making it appear that they all came from the
firewall. Reply packets coming back are translated back,
and forwarded to the appropriate internal machine. Thus,
inside machines are allowed to connect to the outside world.
However, outside machines cannot: in fact, they cannot even *find*
the internal machines, since they are aware of only *one* IP
address, that of the firewall. Thus, they cannot attack
the internal machines directly.
Besides providing this type of basic security, Masquerading
also simplifies network administration: The admin of the
internal network can choose reserved IP addresses, e.g.
in the 10.x.x.x range, or the 192.168.x.x range. These
addresses do not have to be registered with the InterNIC,
and can be used however the sysadmin wants, as long as they
are not used on the external network. Note that this also
alleviates the shortage of IP addresses that ISP's are facing:
A site with hundreds of computers can get by with a mere 8
or 16 Internet IP addresses, without denying any of it's users
- Interactive Web Site Security
- An increasing number of web sites are becoming interactive
by having cgi-bins or Java applets that access some database
or other service. However, this sort of access can be a
big security problem: the database typically has to be hidden
behind a firewall, where it cannot be attacked, the web server
and cgi-bins/Java applets must obviously be outside the firewall,
so that web surfers can get to it. This is particularly true
if the database contains customer information, financial
information or other sensitive, confidential information, or if
the database runs on a mainframe or other internal server that cannot
or should not be connected directly to the Internet.
NAT in the form of Port Forwarding can provide an almost ideal
solution to this access problem. On the firewall, IP packets
that come in to a specific port number can be re-written and
forwarded to the internal server providing the actual service.
The reply packets from the internal server are re-written to
make it appear that they came from the firewall. Thus, Port
Forwarding is becoming, and will be a very important Internet
- Mobile Employees
- An increasing number of corporate employees have gone mobile:
they are roving about, with their laptops, doing work at
customer locations. However, these employees need access
to internal servers, and they need that access to be
secure and encrypted. One solution for such access
requirements is to run and encryption technology, such
as SKIP or IPSEC to the firewall, with a configuration
that gives users access to important internal servers and
Currently, the leading IPSEC implementation on Linux is
FreeSwan; it co-exists
just fine with the standard Linux iptables network filtering
code. Note that there are IPsec clients available for many
versions of Windows, and that Microsoft ships an IPsec
implementation with Windows XP, although the XP license may
prevent you from legally using this in interesting ways;
in particular, using it in together with VNC.
Alternately, mobile employees may want to publish servers
on thier laptops, and make those servers fidable and available
despite a dynamically asssigned IP address. Technology that
enables mobile IP which allowing servers on the mobile platform
is refered to as RAT -- Reverse Address Transalation.
The last major revision to this page was in 1998, and
many of the links and references on this page are rather outdated.
Things have moved on since this page was first written.
A fairly flexible and
sophisticated NAT has been implemented in the 2.4.x Linux kernel,
through a set of highly flexible IP filter tables. The
iptables utility is the prefered way of configuring
Linux network translation. You should check to see if iptables
meets your needs; if it does not, then you should probably
investigate some of the references below.
Alternatives, Hints & Solutions
- Netfilter/IP Tables
is the defacto standard NAT/packet-filtering/firewall tool
for Linux-2.4 and later kernels. Chances are excellent that
your favorite Linux distribution has
packet-filtering/firewalling correctly enabled in the default
kernel, and includes the iptables utilities as a separately
installable package. Thus, you need merely to install the
tools, and then read/understand the
FAQ's, HOWTO's and Tutorials. There are also several
graphical tools for configuring the filter rules; however,
they always seemed to be a bit underwhelming, failing to
significantly simplify the (somewhat arduous and complex)
task of setting up the filter rules.
- IPChains is an older Linux firewall/packet-filtering tool for
Linux-2.2 kernels. It has been replaced by IP Tables (above).
is a standard network utility that should come with all Linux
distributions, and so the place to search for documentation is
on your own computer: man ipchains. See also the
ipchains HOWTO. ipchains replaces the older iptables and
ipfw utilities. If you are running the older Linux 2.0 or 2.2
kernels, then the
Masquerade HOWTO is the aproriate place to start.
Those interested in NAT for firewall and security purposes should
Linux Firewall Tools web page.
- The Eddie Project
Eddie Project offers
a broad and powerful set of Open Source tools for solving a variety
of cluster management and server farm load balancing problems.
This project, supported by the telecom giant Ericsson, appears to
be the most comprehensive, well-balanced package of offerings out
there. It provides support for four major subsystems:
- The IP Migration
Appliation is a general set of cluster management and failover
Load-Balancing DNS Server is a replacement for BIND
that will answer DNS queries by drawing IP's from a pool of
- The Intelligent
HTTP Gateway will redistribute HTTP requests based on rules
about the particular URL as well as load-based algorithms.
- The Content
Replication Application is a recently-started development project.
Its goal is to provide a robust, fault-tolerant, replicated,
geographically distributed file system.
Users with experience with and opinions about
DFS or AFS are encouraged to contact the developers. Note that Linux
is the primary development platform for Coda.
There are two free implementatons of AFS:
There is also Transarc support for
commercial AFS clients and servers.
DCE is in beta test
and a DFS port is in discussion.
InterMezzo is another
distributed file system for Linux.
- RFC 1631
describes the "traditional" NAT (Network Address
Translation) that can be used for this kind of a task.
Basically, the idea behind NAT is to re-write the IP headers
and substitute one numeric address for another. This document
discusses some basic implementation issues, such as computing
header checksums, and mentions problems with packet encryption,
and ICMP. It does not discuss load-balancing or masquerading
Some limitations of this traditional approach are discussed in the
Linux IP NAT theory of operation, including masquerading,
load-balancing, fragmentation and keeping kernel state information.
- One variation of NAT, called masquerading, is already
available in stock Linux kernels. The theory, tools and
installation procedure are discussed in the
IP Masquerade mini-HOWTO. Masquerading is designed to provide
security. It is
intended for use as a type of a firewall, hiding many hosts behind
one IP address, and relabeling all packets from behind the firewall
so that they appear to be coming from on location, the firewall
itself. IP Masq is very powerful and flexible in this respect,
and the filter & accounting rules can configured to handle complex
network topologies. However, it does not currently support
the inverse operation of distributing incoming packets to multiple
Note that Linux Masq does not only "pure" NAT, (i.e. not only
re-writing IP packet headers), but also "impure" packet re-writing
in order to handle the use of services such as FTP, IRC, quake,
RealAudio, CUSeeMe, VDO Live, Microsoft PPTP, etc. from behind the
- Linux IP Network Address Translation
- Currently, there are more NAT implementations for Linux than one can
shake a stick at. They vary in features supported, design choices,
popular appeal, and more. Personally, I beleive that the situation
is ripe for consolidation and collaboration.
NAPT (Network Address and Port Translation)
in general is becoming a *very* important web
technology, since more and more web sites are trying to be
interactive. To be interactive, they have to have cgi-bins or
Java applets or whatever that have to chat with one or more
databases. The database may contain things like customer info,
credit card numbers, or other financial/personal/confidential
stuff, and so you want to hide it behind the firewall. NAPT
is *crucial* for allowing client access while
maintaining security. (The firewall, since it presents a
single point of entry, is easier to monitor and guard than
trying to have many machines exposed on the Internet).
- The earliest version was developed by
Michael Hasenstein, and provides kernel extensions and the
ipnatadm administration tool. This implementation is
maintained, and discussed at
Linux IP Network Address Translation site.
- Another prominent implementation, by Wensong Zhang, aims to work
around a variety of problems with the original implementation,
and is specifically aimed at providing support for clustered,
distributed operation (about which, see the Linux Virtual Server
Wensong's implementation also aspires to wed more closely with
the existing Linux Masq code to provide a more unified approach
to Linux NAT in general.
- IP Filter is a
FreeBSD/NetBSD/OpenBSD implementation but it also works with Linux.
It provides a number of user-level tools, the most interesting
of which is
FLC, a filter rule compiler. If you've ever written filter
rules for a firewal or a router, you will understand how difficult
tedious a task this is. FLC offers a simpler, more flexible way
of creating such rules. It should work with most of the Linux
packaages, as well as with some Cisco products.
- Micheal provides a nice paper titled
"Linux IP NAT Theory of Operation". It includes a discussion of
traditional NAT, NAT as used for masquerading and for
load-balancing, and a discussion of implementation issues.
- Linux Port Forwarding.
Another variation of NAT has been implemented by Steven Clarke,
and is called
Port Forwarding. This is a patch against the masquerading code
in recent Linux kernels that allows IP traffic to certain ports
on the masquerading firewall to be forwarded to internal hosts.
As in masquerading, the IP addresses are re-written so that incoming
IP (TCP and UDP) packets are forwarded, and the reply packets
are rewritten to appear to be coming from the firewall.
As far as I know, no one has tried any of these NAPT
implementations with ENskip or IPSEC.
Note that this is another important application: with ENskip,
clients can talk in an encrypted fashion with the firewall.
After decoding on the firewall, it would be nice to forward the
packets to the appropriate service behind the firewall.
Anyone who has tried port forwarding with ENskip or IPSEC,
please let me know.
- Linux Virtual Server
Linux Virtual Server Project aims to build a scalable virtual
server from a cluster of real servers by using IP traffic load
balancing mechanisms. The virtual server is implemented as a kernel
module, based on the Linux IP masquerading code and
Steven Clarke's port forwarding code. It can dynamically forward
an arbitrary IP connection on given port on the firewall to a
server choosen from a cluster. Dispatch uses a weighted
round-robin scheduling algorithm.
The goal of this technology is to enable scalable servers, such as
scalable web servers, to be built from a cluster real servers,
while providing the security, filtering rules, and IP hiding/translation
aspects of a NAT-style firewall. In this sense, it provides
a far greater level of masking/hiding that the IBM Network Dispatcher,
and resembles Cisco LocalRedirector more closely in operation.
ONE-IP Project implements network clustering using techniques
that in many ways are superior to traditional NAT as described above.
One method of acheiving distribution is with packets routed
to a gateway which then dispatches based on hardware address, rather
than IP address. Thus, all servers on the LAN segment have the same IP
address, and reply to clients with that single IP address, thus
avoiding the overhead of NAT re-writing. Furthermore, since dispatching
is stateless, and the router, gateway and servers sit on the same segment,
failover of the dispatcher is considerably simplified. Another method
elminates the need for a dispatcher by broadcasting on the local segment,
and having servers respond seletively based on a hash of the source
address. Both of these techniques seem to be quite robust to me;
I've bought into the theory. The code is for NetBSD kernels.
- RAT Reverse Address Translation
- RAT or RAPT (Reverse Address and Port Translation) allows a host whose
real IP address is changing from time to time to remain reachable as
a server via a fixed home IP address. In principle, this should
allow setting up servers on DHCP-run networks. While not a perfect mobility
solution, RAPT together with upcoming protocols like DHCP-DDNS, it may
end up becoming another useful tool in the network admin's arsenal.
Like most Mobile IP research, RAT is being done on Linux first. The
Mobile Computing Group
at the National University of Singapore has a number of Mobile IP projects
underway, including an early stage GPL'ed implementation of RAPT called
and developers are solicited.
- HTTP/WWW Load Balancing
- Several URL-based load balancing technologies are generally available
for Linux, either as open source, or as products.
- The Eddie Project
Intelligent HTTP Gateway that dispatches URL requests
to a pool of web servers. It can analyze incoming URL requests
in order to selectively dispatch them to heterogeneous
web servers. It also includes several different load-balancing
algorithms. Works with Apache and other web servers.
Jamming.Net Server to Balance WWW Loads is a collection of
Java applications for redistributing URL requests to multiple
servers. Includes a nice GUI interface for configuring the system.
Note that this tool is only for web requests, and not general IP
pWEB Parallel Web Server Harness will distribute URL requests
to multiple servers based on load and/or URL, for load balancing or
I/O balancing. The harness can be used with most web servers.
Note that this harness is only for web requests, and not general IP
- Redundant, Load Balanced Firewalls
- A product review entitled
A Solution to Redundant, Load Balanced Firewall Systems
discusses some of the issues surrounding fault tolerance
in relation to firewalls.
- Coyote Point
- Coyote Point Systems, Inc.
offers a EIA rack mount box that provides fault tolerant
load balancing of IP traffic.
- RFC 1794 (DNS Support for Load Balancing)
provides a description of a way of answering
DNS requests for a single, well-known domain name with
multiple, different IP addresses. This allows connections
to a single domain to be routed to geographically separate
servers, without forcing all IP traffic through one NAT-enabled
router. As such, it is geographically robust load-balancing
solution. However, the mechanics of doing the actual load
balancing are considerable more difficult.
A description from the net (edited):
>Date: Wed, 4 Sep 96 17:34:13 CDT
>From: email@example.com (Subburajan Ponnuswamy)
>Subject: Re: Parallel IP Routing
>When you look for a machine
>over the Internet, say www.xxx.com, DNS resolves this name by
>contacting other DNS servers (unless it's already in its cache).
>However, at the other end, www.xxx.com may have multiple IP addresses
>(i.e multiple machines, multiple interfaces, etc). Whenever a request
>comes in for an address with multiple IPs, the DNS server sends all the
>valid IPs back in the reply. However, the order of the IP addresses
>play an important role as the requesting machine will start from the
>IP address which appears first..
>You can implement your own algorithm, to shuffle the addresses,
>add/delete the addresses, etc., whenever a query comes in for
>www.xxx.com. Your algorithm may send IP addresses back to the
>requesting machine, such that the least loaded or geographically
>closest machine appears first. Or you can do a simple round
>robin of IP addresses. The TTL value and domain update interval
>should be set accordingly.
>The recent release of BIND has the code for this feature (very
>simple extension of current named) under contrib.
>--Subbu* -- firstname.lastname@example.org ------ Opinions are mine only..
One serious limitation of DNS-based load balancing is that
many/most implementations of BIND view TTL's of less than
300 seconds, and zone refreshes more often than 15 minutes
as being irrational and the symptom of a broken configuration.
Given that most HTTP transfers only last a few seconds, and
socket keep-alives for minutes, it is impossible to do load balancing
on a fine grain. At best, DNS can support only a "stochastic"
load balancing, redirecting clients to servers randomly, as
various caches in various resolvers expire at random (although small)
- Address Failover, HA
- A related topic occurs in discussions of High Availability (HA)
which deals with issues of fault-tolerance and redundancy. One important
componenet of this is address takeover, where the backup machine takes
over the IP address(es) of the failed machine.
- Fake provides
several failover tools, including IP takeover using ARP spoofing.
The IP can either be a physical or an aliased IP.
- Linux-HA discusses
HA in general and points to additional resources.
- Firewall-1 from checkpoint.com
for another simple description & overview of the same concepts.
The product runs on NT and a number of Unixes. Unconfirmed
rumours of a Linux port. Similar to Linux NAT, it is
integrated with a firewall.
- Shock Absorber
- News (Sept 1996): IBM has made a beta-test version of their software
ShockAbsorber, from the
IBM AlphaWorks web site.
It is available for the IBM RS/6000 only, although their FAQ
indicates that it can distribute to other boxes, such as Linux,
If the above URL's do not work for you, try entering the alpha works
site, and search for "shock-absorber". That site uses funky URL's
that are not valid when put into bookmarks or web pages ...
- Firewall FAQ
- Allot sells routers with built
in load-balancing and IP-accounting features.
Copyright (c) 1996, 1997, 1998, 2002 Linas Vepstas All Rights Reserved.
Created 31 August 1996 -- Linas Vepstas
Last updated November 2002 -- by Linas Vepstas
Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free Documentation License, Version 1.1;
with no Invariant Sections, with no Front-Cover Texts, and with no
Back-Cover Texts. A copy of the license is included at the URL
the web page titled
"GNU Free Documentation License".
Go Back to the Enterprise Linux(TM) page
Go Back to Linas' Home page