Tech-invite3GPPspaceIETFspace
96959493929190898887868584838281807978777675747372717069686766656463626160595857565554535251504948474645444342414039383736353433323130292827262524232221201918171615141312111009080706050403020100
in Index   Prev   Next

RFC 2194

Review of Roaming Implementations

Pages: 35
Informational

ToP   noToC   RFC2194 - Page 1
Network Working Group                                         B. Aboba
Request for Comments: 2194                                   Microsoft
Category: Informational                                          J. Lu
                                                        AimQuest Corp.
                                                              J. Alsop
                                                       i-Pass Alliance
                                                               J. Ding
                                                              Asiainfo
                                                               W. Wang
                                                   Merit Network, Inc.
                                                        September 1997


                   Review of Roaming Implementations

1.  Status of this Memo

   This memo provides information for the Internet community.  This memo
   does not specify an Internet standard of any kind.  Distribution of
   this memo is unlimited.

2.  Abstract

   This document reviews the design and functionality of existing
   roaming implementations.  "Roaming capability" may be loosely defined
   as the ability to use any one of multiple Internet service providers
   (ISPs), while maintaining a formal, customer-vendor relationship with
   only one.  Examples of cases where roaming capability might be
   required include ISP "confederations" and ISP-provided corporate
   network access support.

3.  Introduction

   Considerable interest has arisen recently in a set of features that
   fit within the general category of "roaming capability" for Internet
   users.  Interested parties have included:

      Regional Internet Service Providers (ISPs) operating within a
      particular state or province, looking to combine their efforts
      with those of other regional providers to offer service over a
      wider area.

      National ISPs wishing to combine their operations with those of
      one or more ISPs in another nation to offer more comprehensive
      service in a group of countries or on a continent.

      Businesses desiring to offer their employees a comprehensive
      package of access services on a global basis.  Those services may
ToP   noToC   RFC2194 - Page 2
      include Internet access as well as secure access to corporate
      intranets via a Virtual Private Network (VPN), enabled by
      tunneling protocols such as PPTP, L2F, or L2TP.

   What is required to provide roaming capability?  The following list
   is a first cut at defining the requirements for successful roaming
   among an arbitrary set of ISPs:

      Phone number presentation
      Phone number exchange
      Phone book compilation
      Phone book update
      Connection management
      Authentication
      NAS Configuration/Authorization
      Address assignment and routing
      Security
      Accounting

   In this document we review existing roaming implementations,
   describing their functionality within this framework.  In addition to
   full fledged roaming implementations, we will also review
   implementations that, while not meeting the strict definition of
   roaming, address several of these problem elements. These
   implementations typically fall into the category of shared use
   networks or non-IP dialup networks.

3.1.  Terminology

   This document frequently uses the following terms:


   home ISP  This is the Internet service provider with whom the user
          maintains an account relationship.


   local ISP This is the Internet service provider whom the user calls
          in order to get access. Where roaming is implemented the local
          ISP may be different from the home ISP.


   phone book
          This is a database or document containing data pertaining to
          dialup access, including phone numbers and any associated
          attributes.
ToP   noToC   RFC2194 - Page 3
   shared use network
          This is an IP dialup network whose use is shared by two or
          more organizations.  Shared use networks typically implement
          distributed authentication and accounting in order to
          facilitate the relationship among the sharing parties. Since
          these facilities are also required for implementation of
          roaming, implementation of shared use is frequently a first
          step toward development of roaming capabilities.  In fact, one
          of the ways by which a provider may offer roaming service is
          to conclude shared use agreements with multiple networks.
          However, to date the ability to accomplish this has been
          hampered by lack of interoperability among shared use
          implementations.

   non-IP dialup network
          This is a dialup network providing user access to the member
          systems via protocols other than IP.  These networks may
          implement phone book synchronization facilities, in order to
          provide systems, administrators and users with a current list
          of participating systems.  Examples of non-IP dialup networks
          supporting phone book synchronization include FidoNet and
          WWIVnet.

4.  Global Reach Internet Consortium (GRIC)

   Led by a US-based Internet technology developer, AimQuest
   Corporation, ten Internet Service Providers (ISPs) from the USA,
   Australia, China, Japan, Hong Kong, Malaysia, Singapore, Taiwan, and
   Thailand formed the Global Reach Internet Connection (GRIC) in May,
   1996.  The goals of GRIC were to facilitate the implementation of a
   global roaming service and to coordinate billing and settlement among
   the membership.  Commercial operation began in December of 1996, and
   GRIC has grown to over 100 major ISPs and Telcos from all over the
   world, including NETCOM, USA; KDD and Mitsubishi, Japan; iStar,
   Canada; Easynet, UK; Connect.com, Australia; Iprolink, Switzerland;
   Singapore Telecom; Chunghwa Telecom, Taiwan; and Telekom Malaysia.
   Information on GRIC is available from http://www.gric.net/.

   In implementing their roaming service, GRIC members have chosen
   software developed by AimQuest. AimQuest Corporation's roaming
   implementation comprises the following major components: the
   AimTraveler Authentication Server (AAS), the AimTraveler Routing
   Server (ARS), and the AimQuest Internet Management System (AIMS),
   software designed to facilitate the billing process. Information on
   the AimQuest roaming implementation is available from
   http://www.aimquest.com/.
ToP   noToC   RFC2194 - Page 4
   The AimTraveler Authentication Server (AAS) runs at each member ISP
   location, and handles incoming authentication requests from NAS
   devices and other AASes. The AimTraveler Routing Server (ARS) can run
   anywhere.  A single routing server can be used where centralized
   routing is desired, or multiple routing servers can be run in order
   to increase speed and reliability or to gateway to networks of
   particularly large partners.

   The first version of the AimTraveler software, deployed by AimQuest
   in May, 1996, supported direct authentication between members of the
   roaming consortium, but as GRIC grew, management of the relationships
   between the authentication servers became a problem. In August. 1996,
   AimQuest began development of the AimTraveler Routing Server (ARS) in
   order to improve scalability.

   The routing server is comprised of two elements: The Central
   Accounting Server and the Central Routing Server.  The Central
   Accounting Server collects all the roaming accounting data for
   settlement.  The Central Routing Server manages and maintains
   information on the authentication servers in the roaming consortium.
   Adding, deleting, or updating ISP authentication server information
   (e.g. adding a new member ISP) may be accomplished by editing of a
   configuration file on the Central Routing Server. The configuration
   files of the AimTraveler Authentication Servers do not need to be
   modified.

   The AimTraveler Authentication and Routing Servers are available for
   various UNIX platforms. Versions for Windows NT are under
   development.  The AimTraveler Authentication Server supports both the
   UNIX password file and Kerberos.

   The AimQuest Internet Management System (AIMS) is designed for large
   ISPs who need a centralized management system for all ISP operations,
   including sales, trouble-ticketing, service, and billing.  AIMS
   produces usage and transaction statement reports, and includes a
   settlement module to produce settlement/billing reports for the
   roaming consortium members.  Based on these reports, the providers
   charge their ISP/roaming customers, and pay/settle the roaming
   balance among the providers.  AIMS currently runs on
   Sun/Solaris/Oracle. A version for Windows NT and SQL Server is
   expected to become available in Q4 1996.

4.1.  Phone number presentation

   Currently there are two principal methods by which GRIC users can
   discover available phone numbers: a Web-based directory provided by
   the GRIC secretariat, and a GRIC phone book client on the user PC
   with dialing capability.
ToP   noToC   RFC2194 - Page 5
4.1.1.  Web based directory

   A directory of GRIC phone numbers is available on the GRIC home page,
   http://www.gric.com/.  The list of numbers is arranged by country and
   provider. For each provider within a country, this directory,
   provided in the form of a table, offers the following information:

      Provider address, voice phone and fax
      Customer support phone number
      Provider domain name
      Primary Domain Name Server
      Secondary Domain Name Server
      Dial-up IP Address
      News server
      Web page
      POP phone numbers (i.e. 1-408-366-9000)
      POP locations (i.e. Berkeley)
      Proxy addresses
      Dialer configuration

   In order to discover phone numbers using the Web-based directory, it
   is expected that users will be online, and will navigate to the
   appropriate country and provider. They then look up the number and
   insert it into the AimQuest Ranger dialer.

4.1.2.  GRIC phone book client

   The GRIC phone book client software provides for phone book
   presentation as well as automated updating of phone numbers.  The
   GRIC phone book includes a list of countries, states, cities and
   area/city codes, as well as detailed provider information, including
   the cutomer support phone number, and Internet server configuration
   info.  The Phone book, developed with Java, is available for download
   from the AimQuest Web site:

     http://www.aimquest.com/dialer.html

4.2.  Phone number exchange

   GRIC members submit information both about themselves and their POPs
   to the GRIC secretariat, which is run by AimQuest. The GRIC
   secretariat then compiles a new phone book and provides updates on
   the GRIC FTP and Web servers.

   GRIC users then download the phone numbers either in Windows .ini
   file format or in HTML.
ToP   noToC   RFC2194 - Page 6
4.3.  Phone book compilation

   GRIC phone books are compiled manually, and represent a concatenation
   of available numbers from all the members of the roaming consortium,
   with no policy application.  As new POPs come online, the numbers are
   forwarded to GRIC, which adds them to the phone book servers.

4.4.  Phone book update

   Phone numbers in the GRIC phone book client are updated automatically
   upon connection.  The AimTraveler server includes an address book
   which contains the phone numbers of all the roaming consortium
   members.

4.5.  Connection management

   The AimTraveler software supports SLIP and PPP, as well as PAP and
   CHAP authentication.

4.6.  Authentication

   GRIC implements distributed authentication, utilizing the user's e-
   mail address as the userID (i.e. "liu@Aimnet.com") presented to the
   remote NAS device.

   After the initial PPP authentication exchange, the userID, domain,
   and pasword information (or in the case of CHAP, the challenge and
   the response) are then passed by the NAS to the AimTraveler
   Authentication Server which supports both TACACS+ and RADIUS.

   If the authentication request comes from a regular customer login,
   normal user id and password authentication is performed. If the user
   requesting authentication is a "roamer," (has a userID with an @ and
   domain name), the authentication server sends an query to the closest
   routing server. When AimTraveler Routing Server receives the
   authentication request, it first authenticates the AAS sending the
   request, and if this is successful, it checks its authentication
   server table.  If it is able to match the domain of the user to that
   of a "Home ISP", then the Home ISP authentication server's routing
   information are sent back to the local ISP's authentication server.
   Based on the information received from the routing server, the AAS
   makes an authentication request to the user's Home ISP AAS for user
   id and password verification.

   If the user is a valid user, the Home ISP authentication server sends
   a "permission granted" message back to the Local ISP authentication
   server. The Local ISP authentication server then requests the NAS to
   grant the user a dynamic IP address from its address pool. If the
ToP   noToC   RFC2194 - Page 7
   username or password is incorrect, the Home ISP AAS will send a
   rejection message to the Local ISP AAS, and the user will be dropped
   by the NAS.

   If multiple routing servers are installed, and the query to the first
   routing server does not result in a match, the query is forwarded to
   the next routing server. The server queries are cached on the routing
   servers, improving speed for repeated queries. The cache is sustained
   until a routing server table entry is updated or deleted.  Updating
   or deleting results in a message to all neighbor routing servers to
   delete their caches.

   The local authentication server also receives the accounting data
   from the NAS.  If the data is for a regular customer login, the data
   is written to the Local ISP AAS log file. If the data is for a
   "roamer," the data is written to three places: the Local ISP AAS log
   file, the Home ISP AAS log file, and the ARS log file.

   If the local ISP authentication server has caching turned on, then it
   will cache information on Home ISP authentication server
   configurations sent by the routing server. This means that if the
   same domain is queried again, the local authentication server does
   not need to query the routing server again. The local cache is
   cleared when the local authentication server receives an update
   message from the routing server or system manager.

4.7.  NAS Configuration/Authorization

   AimTraveler is comprised of two components, a Client(AAS) and a
   Server(ARS).

   The AimTraveler Client acts as the PPP dial-up authentication server.
   When it detects an '@' sign in the userID field, it queries the
   AimTraverler Server for routing information, then forwards the
   authentication request to user's home authentication server.  The
   AimTraveler Server, a centralized routing server, contains the
   authorized ISP's domain name, authentication servers and other
   information.

   The AimTraveler currently supports RADIUS and TACACS+, and could be
   extended to support other authentication protocols.  It also receives
   all the accounting records, which are subsequently used as input data
   for billing.

   Since ISPs' NAS devices may be configured differently, the attributes
   returned by the home ISP AAS are discarded.
ToP   noToC   RFC2194 - Page 8
4.8.  Address assignment and routing

   All addresses in GRIC are assigned dynamically from within the
   address pool of the local ISP.  Static addresses and routed LAN
   connections will be considered in the future, when GRIC offers
   corporate roaming service, with the implementation of tunneling
   protocols

4.9.  Security

   The user's password is hashed with MD5 before being sent from the
   Local ISP AAS to the Home ISP AAS.  An encryption key is shared
   between the AAS and ARS. The current version of AimTraveler AAS does
   not support token cards or tunneling protocols.

4.10.  Accounting

   The AimTraveler Authentication Server (AAS) software can act as
   either a RADIUS or TACACS+ accounting server.  When accounting
   information is received from the NAS, the local AimTraveler
   Authentication Server (AAS) sends accounting data (user name, domain
   name, login time) to both the Central Accounting Server (part of the
   ARS) and the user's Home ISP AimTraveler authentication server. In
   the case of GRIC, the Central Accounting Server is run by AimQuest.

   The data sent to the central accounting server and home ISP are
   identical except for the form of user id and time stamp.  For a
   traveler whose home ISP is in the US, but who is traveling in Japan,
   the Local (Japanese) ISP AimTraveler authentication server will
   receive an accounting record timestamped with Japan time while the
   Home (US) ISP AimTraveler authentication server will receive an
   accounting record timestamped with the appropriate US timezone.

   The accounting data includes 2 new attributes for settlement
   reporting:

     Attribute              Number   Type
     ---------              ------   ----

     Roaming-Server-ID       101     string
     Isp-ID                  102     string

   The Roaming-Server-ID attribute identifies the AAS sending the
   authentication request.  The Isp-ID attribute identifies the local
   ISP.  Using this information the home ISP can track the roaming
   activities of its users (where their users are logging in).
ToP   noToC   RFC2194 - Page 9
   The AimTraveler Server running at AimQuest keeps a record of all
   roaming transactions, which are used as input to the settlement and
   billing process.  At the end of each month, AimQuest provides a
   roaming transaction summary to GRIC members using AIMS. The AIMS
   software is configurable so that it takes into account the settlement
   rules agreed to by GRIC members.

5.  i-Pass implementation

5.1.  Overview

   i-Pass Alliance Inc., based in Mountain View, California, has
   developed and operates a commercial authentication and settlement
   clearinghouse service which provides global roaming between Internet
   service providers.  The service is fully operational.

   i-Pass Alliance Inc. has additional offices in Toronto, Singapore,
   and London.  More information on i-Pass can be obtained from
   http://www.ipass.com.

   The i-Pass network consists of a number of servers that provide
   real-time authentication services to partner ISPs.  Authentication
   requests and accounting records for roaming users are encrypted and
   sent to an i-Pass serverwhere they are logged, and then forwarded to
   a home ISP for authentication and/or logging.

   Periodically, i-Pass reconciles all accounting records, generates
   billing statements, and acts as a single point for collecting and
   remitting payments.

   i-Pass provides its service only to ISPs and channel partners.  It
   does not attempt to establish a business relationship with
   individual-user customers of an ISP.

5.2.  Access Point Database (APD)

   i-Pass maintains a list of roaming access points in an Oracle
   database.  This list is searchable by geographical region using a Web
   browser, and may be downloaded in its entirety using FTP. The
   information stored for each access point includes:

      Name of service provider
      Country
      State or Province
      City or Region
      Telephone number
      Technical support phone number
      Service types available
ToP   noToC   RFC2194 - Page 10
      Technical information (help file)
      Service pricing information

   The Access Point Database is maintained by i-Pass staff, based on
   input from i-Pass partners.

5.3.  Phone number presentation

   ni-Pass has developed a Windows application wth a simple point and
   click interface called the "i-Pass Dial Wizard", which assists end-
   users in selecting and connecting to a local Internet access point.

   The Dial Wizard allows users to first select the country in which
   they are roaming.  A list of states, provinces, or other regions in
   the selected country is then presented.  Finally a list of access
   points within the state or province is presented.  The Dial Wizard
   displays the city name, modem phone number, and price information for
   each access point within the state or region.

   When the user selects the desired access point, a Windows 95 "DialUp
   Networking" icon is created for that access point.  If there is a
   login script associated with the access point, the DialUp Scripting
   tool is automatically configured.  This means that end-users never
   have to configure any login scripting requirements.

   The Dial Wizard has a built-in phonebook containing all the i-Pass
   access points.  The phonebook may be automatically refreshed from a
   master copy located onISPs web site.

   The Dial Wizard is provided free of charge to i-Pass partners.  i-
   Pass also provides the i-Pass Dial Wizard Customization Kit which
   allows ISP partners to generate customized versions of the Dial
   Wizard with their own logo, etc.

5.4.  Authentication

   There are three entities involved in servicing an authentication
   request:


   Local ISP  At the local ISP, the authentication server is modified to
          recognize user IDs of the form username@auth_domain as being
          remote authentication requests.  These requests are forwarded
          to an i-Pass server.
ToP   noToC   RFC2194 - Page 11
   i-Pass Server
          The i-Pass server receives the authentication request, logs
          it, and forwards it to the home ISP identified by the
          auth_domain portion of the user ID.

   Home ISP The home ISP receives the authentication request, performs
          authentication using its normal authentication method, and
          returns a YES/NO response to the i-Pass server, which in turn
          forwards the reply to the originating ISP.

   i-Pass provides software components which run on the authentication
   servers of the local and home ISPs.  Each member ISP must integrate
   these components with the native authentication method being used by
   the ISP.  To simplify this task, i-Pass has developed "drop-in"
   interfaces for the most commonly used authentication methods.  At the
   date of writing, the following interfaces are supported:

      Livingston RADIUS
      Ascend RADIUS
      Merit RADIUS
      TACACS+
      Xylogics erpcd (Versions 10 and 11)

   A generic interface is also provided which authenticates based on the
   standard UNIX password file.  This is intended as a starting point
   for ISPs using authentication methods other than those listed above.

   The software integration effort for a typical ISP is on the order of
   2-5 man-days including testing.  Platforms currently supported
   include:

      Solaris 2.5 (Sparc).LI
      Solaris 2.5 (Intel)
      BSDI
      Digital Unix
      Linux
      FreeBSD
      HP/UX

   ISPs may chooe to provide authentication for their end-users roaming
   elsewhere, but not to provide access points to the i-Pass network.
   In this case the software integration effort is greatly reduced and
   can be as little as 1/2 a man-day.

5.5.  Accounting

   Accounting transactions are handled in the same way as authentication
   requests.  In addition to being logged at the i-Pass servers,
ToP   noToC   RFC2194 - Page 12
   accounting transactions are sent in real-time to the home ISP.  This
   is intended to allow ISPs to update users' credit limit information
   on a real-time basis (to the extent that this capability is supported
   by their billing and accounting systems).

   Settlement is performed monthly.  The settlement process involves
   calculating the costs associated with each individual session, and
   aggregating them for each ISP.  A net amount is then calculated which
   is either due from i-Pass to the ISP, or from the ISP to i-Pass,
   depending on the actual usage pattern.

   The following reports are supplied to member ISPs:

      A Monthly Statement showing summaries of usage, service provided,
      and any adjustments along with the net amount owing.

      A Call Detail Report showing roaming usage by the ISP's customers.

      A Service Provided report showing detailed usage of the ISP's
      facilities by remote users.

   The above reports are generated as ASCII documents and are
   distributed to i-Pass partners electronically, either by e-mail or
   from  a  secure area on the i-Pass web site. Hard-copy output is
   available on request.

   The Call Detail Report is also generated as a comma-delimited ASCII
   file suitable for import into the ISP's billing database. The Call
   Detail Report will normally be used by the ISP to generate end-user
   billing for roaming usage.

5.6.  Security

   All  transactions  between  ISPs  and the i-Pass servers are
   encrypted using the SSL protocol.  Public key certificates are
   verified at  both the  client  and  server. i-Pass issues these
   certificates and acts as the Cetificate Authority.

   Transactions are also verified based on a number of other criteria
   such as source IP address.

5.7.  Operations

   i-Pass operates several authentication server sites.  Each site
   consists of two redundant server systems located in secure facilities
   and "close" to the Internet backbone.  The authentication server
   sites are geographically distributed to minimize the possibility of
   failure due to natural disasters etc.
ToP   noToC   RFC2194 - Page 13
   i-Pass maintains a Network Operations Center in Mountain View which
   is staffed on a 7x24 basis.  Its functions include monitoring the i-
   Pass authentication servers, monitoring authentication servers
   located at partner facilities, and dealing with problem reports.

6.  ChinaNet implementation

   ChinaNet, owned by China Telecom, is China's largest Internet
   backbone.  Constructed by Asiainfo, a Dallas based system integration
   company, it has 31 backbone nodes in 31 Chinese provincial capital
   cities.  Each province is building its own provincial network, has
   its own dialup servers, and administers its own user base.

   In order to allow hinaNet users to be able to access nodes outside
   their province while traveling, a nationwide roaming system has been
   implemented.  The roaming system was developed by AsiaInfo, and is
   based on the RADIUS protocol.

6.1.  Phone number presentation

   Since China Telecom uses one phone number (163) for nationwide
   Internet access, most cities have the same Internet access number.
   Therefore a phone book is not currently required for the ChinaNet
   implementation. A web-based phone book will be added in a future
   software release in order to support nationwide ISP/CSP telephone
   numbers and HTTP server addresses.

6.2.  Connection management

   The current roaming client and server supports both PPP and SLIP.


6.3.  Address assignment and routing

   ChinaNet only supports dynamic IP address assignment for roaming
   users. In addition, static addresses are supported for users
   authenticating within their home province.

6.4.  Authentication

   When user accesses a local NAS, it provides its userID either as
   "username" or "username@realm".  The NAS will pass the userID and
   password to the RADIUS proxy/server.  If the "username" notation is
   used, the Radius proxy/server will assume that the user is a local
   user and will handle local authentication accordingly.  If "user-
   name@realm" is used, the RADIUS proxy/server will process it as a
   roaming request.
ToP   noToC   RFC2194 - Page 14
   When the RADIUS proxy/server handles a request from a roaming user,
   it will first check the cache to see if the user information is
   already stored there. If there is a cache hit, the RADIUS
   proxy/server do the local authentication accordingly.  If it does not
   find user information in its cache, it will act as a proxy,
   forwarding the authentication request to the home RADIUS server.
   When the home RADIUS server responds, the local server will forward
   the response to the NAS.  If the user is authenticated by the home
   server, the local RADIUS proxy will cache the user information for a
   period of time (3 days by default).

   Caching is used to avoid frequent proxying of requests and responses
   between the local RADIUS proxy and the home RADIUS server.  When the
   home RADIUS server sends back a valid authentication response, the
   local RADIUS proxy/server will cache the user information for a
   period of time (3 days by default).  When the user next authenticates
   directly against the home RADIUS server, the home RADIUS server will
   send a request to the local server or servers to clear the user's
   information from the cache.

6.4.1.  Extended hierarchy

   In some provinces, the local telecommunications administration
   Provincial ISP) further delegates control to county access nodes,
   creating another level of hierarchy. This is done to improve
   scalability and to avoid having the provincial ISP databases grow too
   large.  In the current implementation, each provincial ISP maintains
   its own central RADIUS server, including information on all users in
   the province, while county nodes maintain distributed RADIUS servers.
   For intra-province roaming requests the local RADIUS proxy/server
   will directly forward the request to the home RADIUS server.

   However, for inter-province roaming requests, the local RADIUS server
   does not forward the request directly to the home RADIUS server.
   Instead, the request is forwarded to the central provincial RADIUS
   server for the home province. This implementation is suitable only
   when county level ISPs do not mind combining and sharing their user
   information.  In this instance, this is acceptable, since all county
   level ISPs are part of China Telecom. In a future release, this
   multi-layer hierarchy will be implemented using multi-layer proxy
   RADIUS, in a manner more resembling DNS.

6.5.  Security

   Encryption is used between the local RADIUS proxy/server and the home
   RADIUS server. Public/Private key encryption will be supported in the
   next release. IP tunneling and token card support is under
   consideration.
ToP   noToC   RFC2194 - Page 15
6.6.  Accounting

   Accounting information is transferred between the local RADIUS
   accounting proxy/server and home RADIUS accounting server.  Every day
   each node sends a summary accounting information record to a central
   server in order to support nationwide settlement. The central server
   is run by the central Data Communication Bureau of China Telecom.
   Every month the central server sends the settlement bill to the
   provincial ISPs.

6.7.  Inter-ISP/CSP roaming

   ChinaNet supports both ISP and CSP (Content Service Provider) roaming
   on its system. For example, Shanghai Online, a Web-based commercial
   content service, uses RADIUS for authentication of ChinaNet users who
   do not have a Shanghai Online account. In order to support this, the
   Shanghai Online servers function as a RADIUS client authenticating
   against the home RADIUS server. When users access a protected
   document on the HTTP server, they are prompted to send a
   username/password for authentication. The user then responds with
   their userID in "user-name@realm" notation.

   A CGI script on the HTTP server then acts as a RADIUS authentication
   client, sending the request to the home RADIUS server. After the home
   RADIUS server responds, the CGI script passes the information to the
   local authentication agent. From this point forward, everything is
   taken care of by the local Web authentication mechanism.

7.  Microsoft implementation

   Microsoft's roaming implementation was originally developed in order
   to support the Microsoft Network (MSN), which now offers Internet
   access in seven countries: US, Canada, France, Germany, UK, Japan,
   and Australia.  In each of these countries, service is offered in
   cooperation with access partners.  Since users are able to connect to
   the access partner networks while maintaining a customer-vendor
   relationship with MSN, this implementation fits within the definition
   of roaming as used in this document.

7.1.  Implementation overview

   The first version of the Microsoft roaming software was deployed by
   the MSN partners in April, 1996.  This version included a Phone Book
   manager tool running under Windows 95, as well as a RADIUS
   server/proxy implementation running under Windows NT; TACACS+ is
ToP   noToC   RFC2194 - Page 16
   currently not supported.  Additional components now under development
   include a Connection Manager client for Windows 95 as well as an
   HTTP-based phone book server for Windows NT. The Phone Book manager
   tool is also being upgraded to provide for more automated phone book
   compilation.


7.2.  Phone number presentation

   The Connection Manager is responsible for the presentation and
   updating of phone numbers, as well as for dialing and making
   connections.  In order to select phone numbers, users are asked to
   select the desired country and region/state.  Phone numbers are then
   presented in the area selected.  The primary numbers are those from
   the users service provider which match the service type (Analog,
   ISDN, Analog & IDN), country and region/state selected. The other
   numbers (selected clicking on the More button) are those for other
   service providers that have a roaming agreement with the users
   service provider.

7.2.1.  Cost data

   Cost data is not presented to users along with the phone numbers.
   However, such information may be made available by other means, such
   as via a Web page.

7.2.2.  Default phone book format

   The Connection Manager supports the ability to customize the phone
   book format, and it is expected that many ISPs will make use of this
   capability. However, for those who wish to use it "off the shelf" a
   default phone book format is provided. The default phone book is
   comprised of several files, including:

      Service profile
      Phone Book
      Region file

   The service profile provides information on a given service, which
   may be an isolated Internet Service Provider, or may represent a
   roaming consortium. The service profile, which is in .ini file
   format, is comprised of the following information:

      The name of the service
      The filename of the service's big icon
      The filename of the service's little icon
      A description of the service
      The service phone book filename
ToP   noToC   RFC2194 - Page 17
      The service phone book version number
      The service regions file
      The URL of the service phone book server
      The prefix used by the service (i.e. "MSN/aboba")
      The suffix or domain used by the service (i.e. "aboba@msn.com")
      Whether the user name is optional for the service
      Whether the password is optional for the service
      Maximum length of the user name for the service
      Maximum length of the password for the service
      Information on service password handling (lowercase, mixed case, etc.)
      Number of redials for this service
      Delay between redials for this service
      References to other service providers that have roaming agreements
      The service profile filenames for each of the references
      Mask and match phone book filters for each of the references
        (these are 32 bit numbers that are applied against the capability
        flags in the phone book)
      The dial-up connection properties configuration
        (this is the DUN connectoid name)

   The phone book file is a comma delimited ASCII file containing the
   following data:

      Unique number identifying a particular record (Index)
      Country ID
      A zero-base index into the region file
      City
      Area code
      Local phone number
      Minimum Speed
      Maximum speed
      Capability Flags:
        Bit 0: 0=Toll, 1=Toll free
        Bit 1: 0=X25, 1=IP
        Bit 2: 0=Analog, 1=No analog support
        Bit 3: 0=no ISDN support, 1=ISDN
        Bit 4: 0
        Bit 5: 0
        Bit 6: 0=No Internet access, 1=Internet access
        Bit 7: 0=No signup access, 1=Signup access
        Bit 8-31: reserved
      The filename of the dialup network file
        (typically refers to a script associated with the number)
ToP   noToC   RFC2194 - Page 18
   A sample phone book file is shown below:

      65031,1,1,Aniston,205,5551212,2400,2400,1,0,myfile
      200255,1,1,Auburn/Opelika,334,5551212,9600,28800,0,10,
      200133,1,1,Birmingham,205,5551212,9600,28800,0,10,
      130,1,1,Birmingham,205,3275411,9600,14400,9,0,yourfile
      65034,1,1,Birmingham,205,3285719,9600,14400,1,0,myfile

7.2.3.  Additional attributes

   As described previously, it is likely that some ISPs will require
   additional phone number attributes or provider information beyond
   that supported in the default phone book format.  Attributes of
   interest may vary between providers, or may arise as a result of the
   introduction of new technologies.  As a result, the set of phone
   number attributes is likely to evolve over time, and extensibility in
   the phone book format is highly desirable.

   For example, in addition to the attributes provided in the default
   phone book, the following additional attributes have been requested
   by customers:

      Multicast support flag
      External/internal flag (to differentiate display between the
           "internal" or "other" list box)
      Priority  (for control of presentation order)
      Modem protocol capabilities (V.34, V.32bis, etc.)
      ISDN protocol capabilities (V.110, V.120, etc.)
      No password flag (for numbers using telephone-based billing)
      Provider name

7.2.4.  Addition of information on providers

   The default phone book does not provide a mechanism for display of
   information on the individual ISPs within the roaming consortium,
   only for the consortium as a whole. For example, the provider icons
   (big and little) are included in the service profile. The service
   description information is expected to contain the customer support
   number.  However, this information cannot be provided on an
   individual basis for each of the members of a roaming consortium.
   Additional information useful on a per-provider basis would include:

      Provider voice phone number
      Provider icon
      Provider fax phone number
      Provider customer support phone number
ToP   noToC   RFC2194 - Page 19
7.3.  Phone number exchange

   Currently phone number exchange is not supported by the phone book
   server. As a result, in the MSN implementation, phone number exchange
   is handled manually.  As new POPs come online, the numbers are
   forwarded to MSN, which tests the numbers and approves them for
   addition to the phone book server. Updated phone books are produced
   and loaded on the phone book server on a weekly basis.

7.4.  Phone book compilation

   The Phone Book Manager tool was created in order to make it easier
   for the access partners to create and update their phone books. It
   supports addition, removal, and editing of phone numbers, generating
   both a new phone book, as well as associated difference files.

   With version 1 of the Phone Book Administration tool, phone books are
   compiled manually, and represent a concatenation of available numbers
   from all partners, with no policy application.  With version 1, the
   updates are prepared by the partners and forwarded to MSN, which
   tests the numbers and approves them for addition to the phone book.
   The updates are then concatenated together to form the global update
   file.

   The new version of the Phone Book Administration tool automates much
   of the phone book compilation process, making it possible for phone
   book compilation to be decentralized with each partner running their
   own phone book server. Partners can then maintain and test their
   individual phone books and post them on their own Phone Book Server.

7.5.  Phone book update

   There is a mechanism to download phone book deltas, as well as to
   download arbitrary executables which can perform more complex update
   processing.  Digital signatures are only used on the downloading of
   executables, since only these represent a security threat - the
   Connection Manager client does not check for digital signatures on
   deltas because bogus deltas can't really cause any harm.


   The Connection Manager updates the phone book each time the user logs
   on.  This is accomplished via an HTTP GET request to the phone book
   server. When the server is examining the request, it can take into
   account things like the OS version on the client, the language on the
   client, the version of Connection Manager on the client, and the
   version of the phone book on the client, in order to determine what
   it wants to send back.
ToP   noToC   RFC2194 - Page 20
   In the GET response, the phone book server responds with the
   difference files necessary to update the client's phone book to the
   latest version. The client then builds the new phone book by
   successively applying these difference files.  This process results
   in the update of the entire phone book, and is simple enough to allow
   it to be easily implemented on a variety of HTTP servers, either as a
   CGI script or (on NT) as an ISAPI DLL.

   The difference files used in the default phone book consist of a
   list of phone book entries, each uniquely identified by their index
   number.  Additions consist of phone book entries with all the
   information filed in;  deletions are signified by entries with all
   entries zeroed out. A sample difference file is shown below:

      65031,1,1,Aniston,205,5551212,2400,2400,1,0,myfile
      200255,1,1,Auburn/Opelika,334,5551212,9600,28800,0,10,
      200133,0,0,0,0,0,0,0,0,0
      130,1,1,Birmingham,205,5551211,9600,14400,9,0,yourfile
      65034,1,1,Birmingham,205,5551210,9600,14400,1,0,myfile


7.6.  Connection management

   The Connection Manager can support any protocol which can be
   configured via use of Windows Dialup Networking, including PPP and
   SLIP running over IP.  The default setting is for the IP address as
   well as the DNS server IP address to be assigned by the NAS. The DNS
   server assignment capability is described in [1].

7.7.  Authentication

   The Connection Manager client and RADIUS proxy/server both support
   suffix style notation (i.e.  "aboba@msn.com"), as well as a prefix
   notation ("MSN/aboba").

   The prefix notation was developed for use with NAS devices with small
   maximum userID lengths.  For these devices the compactness of the
   prefix notation significantly increases the number of characters
   available for the userID field.  However, as an increasing number of
   NAS devices are now supporting 253 octet userIDs (the maximum
   supported by RADIUS) the need for prefix notation is declining.

   After receiving the userID from the Connection Manager client, the
   NAS device passes the userID/domain and password information (or in
   the case of CHAP, the challenge and the response) to the RADIUS
ToP   noToC   RFC2194 - Page 21
   proxy. The RADIUS proxy then checks if the domain is authorized for
   roaming by examining a static configuration file. If the domain is
   authorized, the RADIUS proxy then forwards the request to the
   appropriate RADIUS server. The domain to server mapping is also made
   via a static configuration file.

   While static configuration files work well for small roaming
   consortia, for larger consortia static configuration will become
   tedious.  Potentially more scalable solutions include use of DNS SRV
   records for the domain to RADIUS server mapping.


7.8.  NAS configuration/authorization

   Although the attributes returned by the home RADIUS server may make
   sense to home NAS devices, the local NAS may be configured
   differently, or may be from a different vendor.  As a result, it may
   be necessary for the RADIUS proxy to edit the attribute set returned
   by the home RADIUS server, in order to provide the local NAS with the
   appropriate configuration information.  The editing occurs via
   attribute discard and insertion of attributes by the proxy.

   Alternatively, the home RADIUS server may be configured not to return
   any network-specific attributes, and to allow these to be inserted by
   the local RADIUS proxy.

   Attributes most likely to cause conflicts include:

      Framed-IP-Address Framed-IP-Netmask Framed-Routing Framed-Route
      Filter-Id Vendor-Specific Session-Timeout Idle-Timeout
      Termination-Action

   Conflicts relating to IP address assignment and routing are very
   common.  Where dynamic address assignment is used, an IP address pool
   appropriate for the local NAS can be substituted for the IP address
   pool designated by the home RADIUS server.

   However, not all address conflicts can be resolved by editing.  In
   some cases, (i.e., assignment of a static network address for a LAN)
   it may not be possible for the local NAS to accept the home RADIUS
   server's address assignment, yet the roaming hosts may not be able to
   accept an alternative assignment.

   Filter IDs also pose a problem. It is possible that the local NAS may
   not implement a filter corresponding to that designated by the home
   RADIUS server. Even if an equivalent filter is implemented, in order
   to guarantee correct operation, the proxy's configuration must track
   changes in the filter configurations of each of the members of the
ToP   noToC   RFC2194 - Page 22
   roaming consortium.  In practice this is likely to be unworkable.
   Direct upload of filter configuration is not a solution either,
   because of the wide variation in filter languages supported in
   today's NAS devices.

   Since by definition vendor specific attributes have meaning only to
   devices created by that vendor, use of these attributes is
   problematic within a heterogeneous roaming consortium. While it is
   possible to edit these attributes, or even to discard them or allow
   them to be ignored, this may not always be acceptable. In cases where
   vendor specific attributes relate to security, it may not be
   acceptable for the proxy to modify or discard these attributes; the
   only acceptable action may be for the local NAS to drop the user.
   Unfortunately, RADIUS does not distinguish between mandatory and
   optional attributes, so that there is no way for the proxy to take
   guidance from the server.

   Conflicts over session or idle timeouts may result if since both the
   local and home ISP feel the need to adjust these parameters.  While
   the home ISP may wish to adjust the parameter to match the user's
   software, the local ISP may wish to adjust it to match its own
   service policy. As long as the desired parameters do not differ too
   greatly, a compromise is often possible.

7.9.  Address assignment and routing

   While the Connection Manager software supports both static and
   dynamic address assignment, in the MSN implementation, all addresses
   are dynamically assigned.

   However, selected partners also offer LAN connectivity to their
   customers, usually via static address assignment. However, these
   accounts do not have roaming privileges since no mechanism has been
   put in place for allowing these static routes to move between
   providers.

   Users looking to do LAN roaming between providers are encouraged to
   select a router supporting Network Address Translation (NAT). NAT
   versions implemented in several low-end routers are compatible with
   the dynamic addressing used on MSN, as well as supporting DHCP on the
   LAN side.

7.10.  Security

   The RADIUS proxy/server implementation does not support token cards
   or tunneling protocols.
ToP   noToC   RFC2194 - Page 23
7.11.  Accounting

   In the MSN roaming implementation, the accounting data exchange
   process is specified in terms of an accounting record format, and a
   method by which the records are transferred from the partners to MSN,
   which acts as the settlement agent.  Defining the interaction in
   terms of record formats and transfer protocols implies that the
   partners do not communicate with the settlement agent using NAS
   accounting protocols.  As a result, accounting protocol
   interoperability is not be required.

   However, for this advantage to be fully realized, it is necessary for
   the accounting record format to be extensible.  This makes it more
   likely that the format can be adapted for use with the wide variety
   of accounting protocols in current use (such as SNMP, syslog, RADIUS,
   and TACACS+), as well as future protocols. After all, if the record
   format cannot express the metrics provided by a particular partner's
   accounting protocol, then the record format will not be of much
   usefor a heterogeneous roaming consortium.

7.11.1.  Accounting record format

   The Microsoft RADIUS proxy/server supports the ability to customize
   the accounting record format, and it is expected that some ISPs will
   make use of this capability. However for those who want to use it
   "off the shelf" a default accounting record format is provided. The
   accounting record includes information provided by RADIUS:

      User Name (String; the user's ID, including prefix or suffix)
      NAS IP address (Integer; the IP address of the user's NAS)
      NAS Port (Integer; identifies the physical port on the NAS)
      Service Type (Integer; identifies the service provided to the user)
      NAS Identifier (Integer; unique identifier for the NAS)
      Status Type (Integer; indicates session start and stop,
        as well as accounting on and off)
      Delay Time (Integer; time client has been trying to send)
      Input Octets (Integer; in stop record, octets received from port)
      Output Octets (Integer; in stop record, octets sent to port)
      Session ID (Integer; unique ID linking start and stop records)
      Authentication (Integer; indicates how user was authenticated)
      Session Time (Integer; in stop record, seconds of received service)
      Input Packets (Integer; in stop record, packets received from port)
      Output Packets (Integer; in stop record, packets sent to port)
      Termination Cause (Integer; in stop record, indicates termination cause)
      Multi-Session ID (String; for linking of multiple related sessions)
      Link Count (Integer; number of links up when record was generated)
      NAS Port Type (Integer; indicates async vs. sync ISDN, V.120, etc.)
ToP   noToC   RFC2194 - Page 24
   However, since this default format is not extensible, it cannot
   easily be adapted to protocols other than RADIUS, services other than
   dialup (i.e. dedicated connections) or rated events (i.e.  file
   downloads).  This is a serious limitation, and as a result, customers
   have requested a more general accounting record format.

7.11.2.  Transfer mechanism

   Prior to being transferred, the accounting records are compressed so
   as to save bandwidth.  The transfer of accounting records is handled
   via FTP, with the transfer being initiated by the receiving party,
   rather than by the sending party.  A duplicate set of records is kept
   by the local ISP for verification purposes.

8.  Merit Network Implementation

8.1.  Overview

   MichNet is a regional IP backbone network operated within the state
   of Michigan by Merit Network, Inc., a nonprofit corporation based in
   Ann Arbor, Michigan. Started in 1966, MichNet currently provides
   backbone level Internet connectivity and dial-in IP services to its
   member and affiliate universities, colleges, K-12 schools, libraries,
   government institutions, other nonprofit organizations, and
   commercial business entities.

   As of May 1, 1997, MichNet had 11 members and 405 affiliates.  Its
   shared dial-in service operated 133 sites in Michigan and one in
   Washington, D.C, with 4774 dial-in lines.  Additional dial-in lines
   and sites are being installed daily.

   MichNet also provides national and international dial-in services to
   its members and affiliates through an 800 number and other external
   services contracting with national and global service providers.

   The phone numbers of all MichNet shared dial-in sites are published
   both on the Merit web site and in the MichNet newsletters. Merit also
   provides links to information about the national and international
   service sites through their respective providers' web sites.  Such
   information can be found at http://www.merit.edu/mich-
   net/shared.dialin/.

8.1.1.  MichNet State-Wide Shared Dial-In Services

   Each MichNet shared dial-in service site is owned and maintained by
   either Merit or by a member or affiliate organization. All sites must
   support PPP and Telnet connections.
ToP   noToC   RFC2194 - Page 25
   Each organization participating in the shared dial-in service is
   assigned a realm-name.  Typically the realm-name resembles a fully
   qualified domain name. Users accessing the shared dial-in service
   identify themselves by using a MichNet AccessID which consists of
   their local id concatenated with "@" followed by the realm-name -
   e.g.  user@realm

   Merit operates a set of Authentication, Authorization and Accounting
   (AAA) servers supporting the RADIUS protocol which are called core
   servers.  The core servers support all the dial-in service sites and
   act as proxy servers to other AAA servers running at the
   participating organizations. For security reasons, Merit staff run
   all core servers; in particular, the user password is in the clear
   when the proxy core server decodes an incoming request and then re-
   encodes it and forwards it out again,

   The core servers also enforce a common policy among all dial-in
   servers.  The most important policy is that each provider of access
   must make dial-in ports available to others when the provider's own
   users do not have a need for them. To implement this policy, the
   proxy server distinguishes between realms that are owners and realms
   that are guests.

   One piece of the policy determining whether the provider's
   organization has need of the port, is implemented by having the proxy
   core server track the realms associated with each of the sessions
   connected at a particular huntgroup. If there are few ports available
   (where few is determined by a formula) then guests are denied access.
   Guests are also assigned a time limit and their sessions are
   terminated after some amount of time (currently one hour during prime
   time, two hours during non-prime time).

   The other part of the policy is to limit the number of guests that
   are allowed to connect.  This is done by limiting the number of
   simultaneous guest sessions for realms.  Each realm is allocated a
   number of "simultaneous access tokens" - SATs.  When a guest session
   is authorized the end server for the realm decrements the count of
   available SATs, and when the session is terminated the count of SATs
   is incremented.  A Merit specific attribute is added to the request
   by the core if the session will be a "guest" and will require a SAT.
   The end server must include a reply with an attribute containing the
   name of the "token pool" from which the token for this session is
   taken.  The effect of this is to limit the number of guests connected
   to the network to the total number of tokens allocated to all realms.
ToP   noToC   RFC2194 - Page 26
   Each realm is authenticated and authorized by its own AAA server. The
   proxy core servers forward requests to the appropriate server based
   on a configuration file showing where each realm is to be
   authenticated.  Requests from realms not in the configuration are
   dropped.

   The Merit AAA server software supports this policy.  Merit provides
   this software to member and affiliate organizations. The software is
   designed to work with many existing authentication servers, such as
   Kerberos IV, UNIX password, TACACS, TACACS+, and RADIUS.  This
   enables most institutions to utilize the authentication mechanism
   they have in place.

8.1.2.  MichNet National and International Dial-In Services

   In addition to the MichNet shared dial-in service, Merit also
   provides access from locations outside of Michigan by interconnecting
   with other dial-in services. These services are typically billed by
   connect time. Merit acts as the accounting agent between its member
   and affiliate organizations and the outside service provider.

   The services currently supported are a national 800 number and
   service via the ADP/Autonet dial-in network. Connection with
   IBM/Advantis is being tested, and several other service interconnects
   are being investigated.

   Calls placed by a Merit member/affiliate user to these external
   dial-in services are authenticated by having each of those services
   forward RADIUS authentication requests and accounting messages to a
   Merit proxy core server. The core forwards the requests to the
   member/affiliate server for approval. Session records are logged at
   the Merit core server and at the member/affiliate erver. Merit bills
   members/affiliates monthly, based on processing of the accounting
   logs. The members and affiliates are responsible for rebilling their
   users.

   The Merit AAA software supports the ability to request positive
   confirmation of acceptance of charges, and provides tools for
   accumulating and reporting on use by realm and by user.

8.2.  Authentication and Authorization

   Authentication of a Telnet session is supported using the traditional
   id and password method, with the id being a MichNet AccessID of the
   form user@realm, while a PPP session may be authenticated either
   using an AccessID and password within a script, or using PAP.
   Support for challenge/response authentication mechanisms using EAP is
   under development.
ToP   noToC   RFC2194 - Page 27
   When a user dials into a MichNet shared dial-in port, the NAS sends
   an Access-Request to a core AAA server using the RADIUS protocol.
   First the core server applies any appropriate huntgroup access
   policies to the request. If the Request fails the policy check, an
   Access-Reject is returned to the NAS.  Otherwise, the core server
   forwards it to the user's home authentication server according to the
   user's realm.  The home authentication server authenticates and
   authorizes the access request.  An Access-Accept or Access-Reject is
   sent back to the core server.  If an Access-Accept is sent, the home
   server will create a dial-in session identifier which is unique to
   this session and insert it in a Class attribute in the Access-Accept.
   The core server looks at the request and the response from the home
   server again and decides either to accept or reject the request.
   Finally, the core server sends either an Access-Accept or Access-
   Reject to the NAS.

   When a user dials into a contracted ISP's huntgrup (MichNet National
   and International Service), the ISP sends a RADIUS access request to
   a Merit core server. The rest of the authentication and authorization
   path is the same as in the shared dial-in service, except that no
   huntgroup access policy is applied but a Huntgroup-Service attribute
   is sent to the home authentication server with its value being the
   name of the service, and a copy of the attribute must be returned by
   the home server with a flag appended to the original value to
   indicate a positive authorization of user access to the specified
   service.

   The MichNet shared dial-in service typically requires authorization
   of some sort, for example, a user dialing into a huntgroup as a guest
   must be authorized with a token from the user's realm. Participating
   institutions have control in defining authorization rules.  Currently
   authorization may be done using any combination of the user's group
   status and user's account status. A set of programming interfaces is
   also provided for incorporating new authorization policies.

8.3.  Accounting

   In the Merit AAA server, a session is defined as starting from the
   moment the user connects to the NAS, and ending at the point when the
   user disconnects. During the course of a session, both the core
   server and the home server maintain status information about the
   session.  This allows the AAA servers to apply policies based on the
   current status, e.g. limit guest access by realm to number of
ToP   noToC   RFC2194 - Page 28
   available tokens, or to limit number of simultaneous sessions for a
   given AccessID. Information such as whether the session is for a
   guest, whether it used a token, and other information is included
   with the accounting stop information when it is logged. Merit has
   made enhancements to the RADIUS protocol, that are local to the AAA
   server, to support maintenance of session status information.

   When a user session is successfully authenticated, the NAS sends out
   a RADIUS accounting start request to the core server. The core server
   forwards that request to the user's home server.  The home server
   updates the status of the session and then responds to the core. The
   core server in turn responds to the NAS.  In the accounting Start
   request, a NAS conforming to the RADIUS specification must return the
   Class attribute and value it received in the Access-Accept for the
   session, thus sending back the dial-in session identifier created by
   the session's home server.

   When a user ends a session, an accounting stop request is sent
   through the same path.  the same path.  The dial-in session
   identifier is again returned by the NAS, providing a means of
   uniquely identifying a session.  By configuring the finite state
   machine in each of the AAA servers, any accounting requests may be
   logged by any of the servers where the accounting requests are
   received.

   Because the same session logs are available on every server in the
   path of a session's authorization and accounting message, problems
   with reconciliation of specific sessions may be resolved easily. For
   the shared dial-in service, there are no usage charges.  Merit has
   tools to verify that organizations do not authorize more guest
   sessions than the number of SATs allocated to the organization.  For
   surcharged sessions, Merit sends each organization a summary bill
   each month. Files with detail session records are available for
   problem resolution.  Each organization is responsible for billing its
   own users, and should have the same session records as are collected
   by Merit.

   Merit receives a monthly invoice from other dial-in service providers
   and pays them directly, after first verifying that the charges
   correspond to the session records logged by Merit.

8.4.  Software and Development

   Merit has developed the AAA server software which supports the above
   capabilities initially by modifying the RADIUS server provided by
   Livingston, and later by doing a nearly total rewrite of the software
   to make enhancement and extension of capabilites easier.  Merit makes
   a basic version of its server freely available for noncommercial use.
ToP   noToC   RFC2194 - Page 29
   Merit has started the Merit AAA Server Consortium which consists of
   Merit and a number of NAS vedors, ISPs and server software vendors.
   The consortium supports ongoing development of the Merit AAA server.
   The goal is to build a server that supports proxy as well as end
   server capabilities, that is feature rich, and that interoperates
   with major vendors' NAS products.

   The building block of the Merit AAA server, the
   Authentication/Authorization Transfer Vector (AATV), is a very
   powerful concept that enables the ultimate modularity and flexibility
   of the AAA server. The structure and methods of the AATV model are
   published with all versions of the AAA server.

   Objects for extending the authorization server are also available in
   the enhanced version of the AAA server. Merit is also looking at ways
   to provide a method of extending the AAA server in its executable
   form, to improve the server efficiency and scalability, and to
   provide better monitoring, instrumentation and administration of the
   server.

9.  FidoNet implementation

   Since its birth in 1984, FidoNet has supported phone book
   synchronization among its member nodes, which now number
   approximately 35,000.  As a non-IP dialup network, FidoNet does not
   provide IP services to members, and does not utilize IP-based
   authentication technology.  Instead member nodes offer bulletin-board
   services, including access to mail and conferences known as echoes.

   In order to be able to communicate with each other, FidoNet member
   systems require a sychronized phone book, known as the Nodelist. The
   purpose of the Nodelist is to enable resolution of FidoNet addresses
   (expressed in the form zone:network/node, or 1:161/445) to phone
   numbers.  As a dialup network, FidoNet requires phone numbers in
   order to be deliver mail and conference traffic.

   In order to minimize the effort required in regularly synchronizing a
   phone book of 35,000 entries, the weekly Nodelist updates are
   transmitted as difference files.  These difference files, known as
   the Nodediff, produce the Nodelist for the current week when applied
   to the previous week's Nodelist.  In order to minimize transfer time,
   Nodediffs are compressed prior to transfer.

   Information on FidoNet, as well as FidoNet Technical Standards (FTS)
   documents (including the Nodelist specification) and standards
   proposals are available from the FidoNet archive at
   http://www.fidonet.org/.
ToP   noToC   RFC2194 - Page 30
9.1.  Scaling issues

   With a Nodelist of 35,000 entries, the FidoNet Nodelist is now 3.1 MB
   in size, and the weekly Nodediffs are 175 KB. In compressed form, the
   Nodelist is approximately 1 MB, and the weekly Nodediff is 90 KB. As
   a result, the transfer of the Nodediff takes approximately 45 seconds
   using a 28,800 bps modem.

   In order to improve scalability, the implementation of a domain name
   service approach is examined in [8]. The proposal evisages use of a
   capability analagous to the DNS ISDN record in order to map names to
   phone numbers, coupled with an additional record to provide the
   attributes associated with a given name.

9.2.  Phone number presentation

   While FidoNet member systems perform hone book synchronization, users
   need only know the FidoNet address of the systems they wish to
   contact. As a result users do not need to maintain copies of the
   Nodelist on their own systems. This is similar to the Internet, where
   the DNS takes care of the domain name to IP address mapping, so that
   users do not have to remember IP addresses.

   Nevertheless, FidoNet systems often find it useful to be able to
   present lists of nodes, and as a result, FidoNet Nodelist compilers
   typically produce a representation of the Nodelist that can be
   searched or displayed online, as well as one that is used by the
   system dialer.

9.2.1.  FidoNet Nodelist format

   The FidoNet Nodelist format is documented in detail in [3].  The
   Nodelist file consists of lines of data as well as comment lines,
   which begin with a semi-colon.  The first line of the Nodelist is a
   general interest comment line that includes the date and the day
   number, as well as a 16-bit CRC. The CRC is included so as to allow
   the system assembling the new Nodelist to verify its integrity.

   Each Nodelist data line contains eight comma separated fields:

      Keyword
      Zone/Region/Net/Node number
      Node name
      Location
      Sysop name
      Phone number
      Maximum Baud rate
      Flags (optional)
ToP   noToC   RFC2194 - Page 31
   FidoNet Nodelists are arranged geographically, with systems in the
   same zone, region, and network being grouped together. As a result,
   FidoNet Nodelists do not require a separate regions file. Among other
   things, the keyword field can be used to indicate that a system is
   temporarily out of service.

   Reference [3] discusses Nodelist flags in considerable detail.  Among
   other things, the flags include information on supported modem
   modulation and error correction protocols.  Reference [4] also
   proposes a series of ISDN capability flags, and [5] proposes flags to
   indicate times of system availability.


9.3.  Phone number exchange

   FidoNet coordinators are responsible for maintaining up to date
   information on their networks, regions, and zones. Every week network
   coordinators submit to their regional coordinators updated versions
   of their portions of the Nodelist. The regional coordinators then
   compile the submissions from their network coordinators, and submit
   them to the zone coordinator. The zone coordinators then exchange
   their submissions to produce the new Nodelist. As a result, it is
   possible that the view from different zones may differ at any given
   time.

9.3.1.  The Nodediff

   The format of the Nodediff is discussed in detail in [3]. In
   preparing the Nodediffs, network coordinators may transmit only their
   difference updates, which can be collated to produce the Nodediff
   directly.

   One weakness in the current approach is that there is no security
   applied to the coordinator submissions. This leaves oen the
   possibility of propagation of fraudulent updates. In order to address
   this, [6] proposes addition of a shared secret to the update files.


9.3.2.  Addition of nodes

   In order to apply for allocation of a FidoNet address and membership
   in the Nodelist, systems must demonstrate that they are functioning
   by sending mail to the local network coordinator.  Once the local
   network coordinator receives the application, they then allocate a
   new FidoNet address, and add a Nodelist entry.
ToP   noToC   RFC2194 - Page 32
9.3.3.  Deletion of nodes

   Since FidoNet nodes are required to be functioning during the zone
   mail hour in order to receive mail, and since nodes receive the
   weekly Nodelist from their local network coordinators on a weekly
   basis, there is a built-in mechanism for discovery of non-functional
   nodes.

   Nodes found to be down are reported to the local network coordinator
   and subsequently marked as down within the Nodelist.  Nodes remaining
   down for more than two weeks may be removed from the Nodelist, at the
   discretion of the network coordinator.

9.4.  Phone book update

   The Nodelist contains the phone numbers and associated attributes of
   each participating system. New Nodelists become available on Fridays,
   and are made available to participating systems by their local
   network coordinators, who in turn receive them from the regional and
   zone coordinators.

   While it is standard practice for participating systems to get their
   Nodelists from their local network coordinators, should the local
   network coordinator not be available for some reason, either the
   updates or the complete Nodelist may be picked up from other network,
   or regional coordinators. Please note that since the view from
   different zones may differ, nodes wishing to update their Nodelists
   should not contact systems from outside their zone.

9.5.  Phone book compilation

   Once FidoNet systems have received the Nodediff, the apply it to the
   previous week's Nodelist in order to prepare a new Nodelist.  In
   order to receive Nodediffs and compile the Nodelist, the following
   software is required:

      A FidoNet-compatible mailer implementation, used to transfer files
      A Nodelist compiler

   One of the purposes of the Nodelist compiler is to apply Nodediffs to
   the previous Nodelist in order to produce an updated Nodelist.  The
   other purpose is to compile the updated Nodelist into the format
   required by the particular mailer implementation used by the member
   system.  It is important to note that while the Nodelist and Nodediff
   formats are standardized (FTS-0005), as is the file transfer protocol
   (FTS-0001), the compiled format used by each mailer is implementation
   dependent.
ToP   noToC   RFC2194 - Page 33
   One reason that compiled formats to differ is the addition of out of
   band information to the Nodelist during the compilation process.
   Added information includes phone call costs as well as shared
   secrets.

9.5.1.  Cost data

   Although cost information is not part of the Nodelist, in compiling
   the Nodelist into the format used by the mailer, Nodelist compilers
   support the addition of cost information. This information is then
   subsequently used to guide mailer behavior.

   Since phone call costs depend on the rates charged by the local phone
   company, this information is local in nature and is typically entered
   into the Nodelist compiler's configuration file by the system
   administrator.

9.5.2.  Shared secrets

   In FidoNet, shared secrets are used for authenticated sessions
   between systems.  Such authenticated sessions are particularly
   important between the local, regional and zone coordinators who
   handle preparation and transmission of the Nodediffs. A single shared
   secret is used per system.

9.6.  Accounting

   Within FidoNet, the need for accounting arises primarily from the
   need of local, regional and zone coordinators to be reimbursed for
   their expenses.  In order to support this, utilities have been
   developed to account for network usage at the system level according
   to various metrics.  However, the accounting techniques are not
   applied at the user level. Distributed authentication and acounting
   are not implemented and therefore users may not roam between systems.

10.  Acknowledgements

   Thanks to Glen Zorn of Microsoft and Lynn Liu and Tao Wang of
   AimQuest for useful discussions of this problem space.

Security Considerations

   Security issues are discussed in sections 5.6 and 6.5.
ToP   noToC   RFC2194 - Page 34
11.  References

   [1]  Cobb, S., "PPP Internet Protocol Control Protocol Extensions for
   Name Server Addresses", RFC 1877, Microsoft, December 1995.

   [2]  Fielding, R., et al., "Hypertext Transfer Protocol - HTTP/1.1.",
   RFC 2068, UC Irvine, January, 1997.

   [3]  Baker, B., R. Moore,  D.  Nugent.   "The  Distribution
   Nodelist." FTS-0005, February, 1996.

   [4]  Lentz, A.  "ISDN Nodelist flags." FSC-0091, June, 1996.

   [5]  Thomas, D. J.  "A Proposed Nodelist flag indicating Online Times
   of a Node." FSC-0062, April, 1996.

   [6]  Kolin, L.   "Security  Passwords  in  Nodelist  Update  Files."
   FSC-0055, March, 1991.

   [7]  Gwinn, R.,  D.  Dodell.  "Nodelist Flag Changes Draft Document."
   FSC-0009, November, 1987.

   [8]  Heller, R.  "A Proposal  for  A  FidoNet  Domain  Name
   Service." FSC-0069, December, 1992.

   [9]  Rigney, C., Rubens, A., Simpson, W., and S. Willens, "Remote
   Authentication Dial In User Service (RADIUS)", RFC 2058, Livingston,
   Merit, Daydreamer, January 1997.

   [10] Rigney, C., "RADIUS Accounting", RFC 2059, Livingston, January
   1997.
ToP   noToC   RFC2194 - Page 35
12.  Authors' Addresses

     Bernard Aboba
     Microsoft Corporation
     One Microsoft Way
     Redmond, WA 98052

     Phone: 206-936-6605
     EMail: bernarda@microsoft.com

     Juan Lu
     AimQuest Corporation
     1381 McCarthy Blvd.
     Milpitas, California 95035

     Phone: 408-273-2730  ext. 2762
     EMail: juanlu@aimnet.net


     John Alsop
     i-Pass Alliance Inc.
     650 Castro St., Suite 280
     Mountain View, CA 94041

     Phone: 415-968-2200
     Fax:   415-968-2266
     EMail: jalsop@ipass.com

     James Ding
     Asiainfo
     One Galleria Tower
     13355 Noel Road, #1340
     Dallas, TX 75240

     Phone: 214-788-4141
     Fax:   214-788-0729
     EMail: ding@bjai.asiainfo.com

     Wei Wang
     Merit Network, Inc.
     4251 Plymouth Rd., Suite C
     Ann Arbor, MI 48105-2785

     Phone: 313-764-2874
     EMail: weiwang@merit.edu