[Search] [txt|pdf|bibtex] [Tracker] [WG] [Email] [Nits]

Versions: 00                                                            
Network Working Group                           Philip J. Nessser II
draft-ietf-pier-applications-00.txt       Nesser & Nesser Consulting
Internet Draft                                            March 1996
Expire in six months

                    IP Addresses in Applications

Status of this Memo

   This document is an Internet Draft. Internet Drafts are working
   documents of the Internet Engineering Task Force (IETF), its Areas,
   and its Working Groups. Note that other groups may also distribute
   working documents as Internet Drafts.

   Internet Drafts are draft documents valid for a maximum of six
   months. Internet Drafts may be updated, replaced, or obsoleted by
   Internet Drafts are draft documents valid for a maximum of six
   months. Internet Drafts may be updated, replaced, or obsoleted by
   other documents at any time. It is not appropriate to use Internet
   Drafts as reference material or to cite them other than as a "working
   draft" or "work in progress".

   Please check the I-D abstract listing contained in each Internet
   Draft directory to learn the current status of this or any other
   Internet Draft.

1.0 Abstract

The Procedures for Internet/Enterprise Renumbering (PIER) Working Group
of the Internet Engineering Task Force (IETF) has been tasked with the
creation of documents to aid renumbering efforts.  This document
defines a series of classes of IP address locations.  Each class
will be described in a general sense, while specific examples
are provided as possible.

2.0 Introduction

In an effort to aid organizations in the efforts of renumbering
the PIER group is producing a series of informational documents
about renumbering.  While much press has been given to the
special case of required renumbering when changing Internet
Service Providers (ISPs), there are many reasons which
necessitate renumbering.  For a very detailed discussion of
the reasons for renumbering the reader is refered to RFC XXXX.
A few of those reasons are given below:

        o Changing ISP's;
        o Company Splitting into smaller subdivisions;
        o Two companies merging;
        o Moving facilities whose physical layout require
          topological changes;
        o Changes in network topology;
        o Moving from a bridged to a routed network;

When developing a renumbering plan the administrator typically
identifies the individual elements on the network and tries to
group those devices into like groups.  A relatively natural
set of groupings is presented below:

        o Routers;
        o Infrastructure Devices: Bridges, Terminal Servers,
          Gateways, Firewalls;
        o Applications Servers:  DNS, Mailhubs, News Servers,
          FTP Servers, WWW Servers, Network Management Systems;
        o End User Systems.

A complete guide to router renumbering has been published as RFC
XXXX, and hence will not be covered in this document.  This
documents will attempt to cover all of the remaining devices.
There will not be a linear mapping between the groups above and
the the classes presented below.  For example, when renumbering
a WWW server, it will be necessary to consider both the
underlying system and the WWW server software.

3.0 Basics

When approaching a renumbering project there are several preliminary
steps which should be addressed before proceding with the project.
Some of the steps may seem unnecessary or overcautious but in the
end they almost always save substantial time and wasted effort.

3.1 Identify the Scope of the Renumbering Project.

It is not sufficient to decide to renumber a series of networks.  Each
network, each device on that network, and each network to which it
connects must be inventoried and designated, since there is
interaction between all three groups.  The last group is especially
important in cases of firewalls,  access control lists, etc.

3.2 Identify the Numeric Boundries of the Renumbering Plan.

It is important to know the ranges of IP addresses which are in use
and will be in use after the renumbering plan is complete.  The
simplest case is renumbering one block of addresses to another block
of addresses.  A more complex case might involve renumbering a block
of addresses into the same block of addresses with a different network
toplogy.  In this case, the order of implementation is critical to a
sucessful project.

3.3 Identify the New Network Topology.

While many cases of renumbering are simply a change of prefix, many
involve a network topology change as well.  For example, moving to a
different building may require a shift in both addresses and
topology.  In the case of a changing topology, the new plan should be
in place before any changes are made.

3.4 Identify the Components.

Once the above steps are completed the basic inventory of devices
identified above can be fleshed out.  Each operating system and
hardware platform has distinctions which may aid or hinder its ability
to renumber gracefully.  This RFC attempts to identify those strenghts
and weaknesses.  For example, some routers and host operating systems
allow multiple IP addresses to be assigned to the same interface.
This is a great boon when renumbering since the old and the new
addresses can both be functioning at the same time.

3.5 Create a Map.

It will be greatly beneficial to create a IP address map between the
old addresses and the new addresses.  This will aid in the migration
in countless ways, as well as serve as a bridge backwards to any need
to translate historical data to the current topology.

3.6 Make the Most of the Pain.

There is no doubt that renumbering IP networks can be a painful process
for all concerned.  Since you have to endure the pain, it is clear
that it would be desireable to make the most of the situation.  Try
and take this opportunity to put as much planning and resources into
the project as possible.  Updating operating systems, hardware
firmware and BIOS'es, new cabling, new equipement, new address
assignment plans, etc. are all items which many be necessary when
renumbering, so place careful emphasis on designing a new system which
is flexible to change and well designed.

4.0 Classes

Each of the following sections will be devoted to one "Class" of
renumbering problem.  These will roughly correspond to a set of
locations that contain IP addresses, and some basic ideas on how
to sucessfully renumber those addresses.  Wherever possible a
set of very specific examples will be given.  Every effort has
been made to provide as many examples as possible, but they are
definitely not exhaustive.

4.1 Firewalls (Including Filtering Routers, Proxy Servers,
           Application Layer Gateways (ALGs), and Network
           Address Translators (NATs)

To Be Done

4.2 Network Management Stations (NMS)

NMS's are typically dedicated machines which runs a SNMP application used
to both monitor systems, routers, and other infrastructure, but also to
collect, store, manage and track that data.  In all known cases, NMS's
store this data based on the IP address of the monitored device or
interface.  Because of the typical volume of this data, it is typically
stored in a proprietary database format to save space, and speed access.

To Be Finished

4.3 Software License Servers

To Be Done

4.4 Names Systems (Including Domain Name System (DNS), Windows
               Internet Naming Service (WINS),  Network
               Information Services (NIS) and NIS Plus)

4.5 DHCP/BOOTP Servers

To Be Done

4.6 Client Configurations (Unix varieties, Windows 95, Windows NT,
                       Mac OS 7.5)

The purpose of all of the infrastructure is to allow operation between
host/client computers.  In this context all machines are clients, in
the sense that all systems underlying operating systems (OS's) need to
be renumbered, even if additional steps might be necessary to renumber
higher layer applications.

Part of the difficulty in this problem is the extreme success of the
TCP/IP protocol.  The ability of IP to run over such a large variety
of level two protocol media has encouraged adoption on so many
hardware platforms and operating systems.  The combinations of
hardware, operating system, operating system version, network software
implementation, network software implementation version, layer two
protocol and layer two media provide a vast, if not unlimited number
of possibilities.  This discussion will center on the likely locations
of hard coded IP addresses in the OS.   Some specific examples will be
provided for some of the most common hardware platforms and OS's.

Operating systems typically have IP addresses in a minimum of three
locations.  First make the assumption that this machine does not use
one of the dynamic address protocols (bootp, dhcp, etc).  The first is
the location of the IP address for machine itself.  The second
location is the default gateway for the network to which the machine
is connected.  The third location is an IP address for a machine which
provides name to IP address mapping (there of course are often
multiple name to IP address servers for redundency).  Many popular
implementations do group all three of these IP addresses in a single
location, making changes straightforward.  This is perhaps the
simplest configuration for a machine.  In many sites a large majority
of machines to be renumbered will fall into this category.  The
observation that 90% of the machines to be renumbered will only take
10% of the time and effort, while the last 10% of the machines will
take over 90% of the time and effort.  These simple client machines
will typically be the bulk of the *number* of machines to renumber but
are typically the easiest.

There are, of course, other locations in operating systems where IP
addresses can be found.  These additional locations can be grouped
into two categories.  The first is a local cache of hostnames mapped
to IP addresses, while the second group is usually dependent on
additional applications running on the system.  The use of IP
addresses for both of these purposes followed the philosophy that
addressed changed so rarely that it was more network efficient to hard
code addresses for local hosts that were accessed often, as well as,
addresses of machines that provided well know services, thus saving
many unneccesary name server lookups.  This is not the case in the
Internet of the 1990's, and should therefore no longer be practiced.

In the first category, for example, the local time sharing machines,
or mail hub, or news server, etc. may be stored locally to avoid
"wasted" nameserver lookups.  In the second category, for example,
there are several time syncronization protocols which allow clients to
sync their time and date with a server somewhere on the network.
There is typically a startup file which identifies the time server's
address.  This is perfect example of a situation where an IP address
is not needed, and where a hostname would be better suited.

In the past, it was common for IP addresses to be used in
situations like this where remote servers rarely changed addresses,
thus it was unnecessary to "waste" namesever lookups in these cases.
In the "new" Internet with its quicky growing infrastructure it is an
unwise decision to hardcode such addresses.  In fact, as "smarter"
application servers are deployed which learn network topology and
provide the location to the "best" server for an application, there
may be significant problems if hard coded addresses are used.

Other examples of possible IP address locations which fall into the
second category include:

        o Configuration of remote time syncronization (as above)
        o Configuration of remote printers
        o Configuration of remote filesystem connections
        o Non-default network to subnet mask mapping
        o Configuration of remote font servers

4.6.1 Examples -- Unix --

4.7 Applications (Web Servers)

To Be Done

4.8 Mail Systems

To Be Done

4.9 Netbios over TCP

To Be Done

4.10 System Security Tools (TCP Wrappers, Socks, Xinetd)

To Be Done

4.11 Documentation  (Online and Offline)

Every system has some sort of online documentation.  At the simplist
level it may be self referential documentation (i.e. hosts file), or
it may be elaborite documentation covering each network node and its
associated network interfaces and services, or it may fall somewhere
in the middle.  Nevertheless, it is vital to update this information
as part of the renumbering process.  In some cases, where the
documents are either in ASCII or stored in a database, it will be
relatively easy to automatically convert the information using a
mapping table from old to new addresses.

However,  there are many places where IP addresses are embedded in a
binary document, say a word processor or spreadsheet file, where
significant manual intervention may be required.

An often over looked location when renumbering is your organizations
off-line documentation.  Over the years most organziations have
developed numerous offline documents which could contain IP numbers.
This may be a good time to do a complete scan of all offline
documentation.  Below is a list of examples which should be checked
for hardcoded IP addresses.

  o System setup information
  o Disaster recovery plans
  o End user documentation
  o Network maps
  o Dialin instructions
  o Numbering schemes
  o List of network resources (DNS servers, gateways, Database
    servers, etc.)

5.0 Security Considerations

   Security issues are not discussed in this memo.

6.0 Authors' Addresses

   Philip J. Nesser II
   Nesser & Nesser Consulting
   13501 100th Ave NE, Suite 5202
   Kirkland, WA 98034

   Phone: (206)481-4303
   Email: pjnesser@martigny.ai.mit.edu

7.0 References