Network Working Group C. Weider
Request for Comments: 1309 ANS
FYI: 14 J. Reynolds
ISI
S. Heker
JvNC
March 1992
Technical Overview of Directory Services
Using the X.500 Protocol
Status of this Memo
This memo provides information for the Internet community. It does
not specify an Internet standard. Distribution of this memo is
unlimited.
Abstract
This document is an overview of the X.500 standard for people not
familiar with the technology. It compares and contrasts Directory
Services based on X.500 with several of the other Directory services
currently in use in the Internet. This paper also describes the
status of the standard and provides references for further
information on X.500 implementations and technical information.
A primary purpose of this paper is to illustrate the vast
functionality of the X.500 protocol and to show how it can be used to
provide a global directory for human use, and can support other
applications which would benefit from directory services, such as
main programs.
This FYI RFC is a product of the Directory Information Services
(pilot) Infrastructure Working Group (DISI). A combined effort of
the User Services and the OSI Integration Areas of the Internet
Engineering Task Force (IETF).
As the pace of industry, science, and technological development
quickened over the past century, it became increasingly probable that
someone in a geographically distant location would be trying to solve
the same problems you were trying to solve, or that someone in a
geographically distant location would have some vital information
which impinged on your research or business. The stupendous growth
in the telecommunications industry, from telegraphs to telephones to
computer networks, has alleviated the problem of being able to
DISI Working Group [Page 1]
RFC 1309 Technical Overview of X.500 March 1992
communicate with another person, PROVIDED THAT YOU KNOW HOW TO REACH
THEM.
Thus, along with the expansion of the telecommunications
infrastructure came the development of Directory Services. In this
paper, we will discuss various models of directory services, the
limitations of current models, and some solutions provided by the
X.500 standard to these limitations.
2 MODELS OF DIRECTORY SERVICES
A model many people think of when they hear the words "Directory
Services" is the directory service provided by the local telephone
company. A local telephone company keeps an on-line list of the names
of people with phone service, along with their phone numbers and
their address. This information is available by calling up Directory
Assistance, giving the name and address of the party whose number you
are seeking, and waiting for the operator to search his database. It
is additionally available by looking in a phone book published yearly
on paper.
The phone companies are able to offer this invaluable service because
they administer the pool of phone numbers. However, this service has
some limitations. For instance, you can find someone's number only if
you know their name and the city or location in which they live. If
two or more people have listings for the same name in the same
locality, there is no additional information which with to select the
correct number. In addition, the printed phone book can have
information which is as much as a year out of date, and the phone
company's internal directory can be as much as two weeks out of date.
A third problem is that one actually has to call Directory assistance
in a given area code to get information for that area; one cannot
call a single number consistently.
For businesses which advertise in the Yellow Pages, there is some
additional information stored for each business; unfortunately, that
information is unavailable through Directory Assistance and must be
gleaned from the phone book.
As the Internet is comprised of a vast conglomeration of different
people, computers, and computer networks, with none of the hierarchy
imposed by the phone system on the area codes and exchange prefixes,
any directory service must be able to deal with the fact that the
Internet is not structured; for example, the hosts foo.com and
DISI Working Group [Page 2]
RFC 1309 Technical Overview of X.500 March 1992
v2.foo.com may be on opposite sides of the world, the .edu domain
maps onto an enormous number of organizations, etc. Let's look at a
few of the services currently available on the Internet for directory
type services.
The finger protocol, which has been implemented for UNIX systems and
a small number of other machines, allows one to "finger" a specific
person or username to a host running the protocol. This is invoked by
typing, for example, "finger clw@mazatzal.merit.edu". A certain set
of information is returned, as this example from a UNIX system finger
operation shows, although the output format is not specified by the
protocol:
Login name: clw In real life: Chris Weider
Directory: /usr/clw Shell: /bin/csh
On since Jul 25 09:43:42 4 hours 52 minutes Idle Time
Plan:
Home: 971-5581
where the first three lines of information are taken from the UNIX
operating systems information and the line(s) of information
following the "Plan:" line are taken from a file named .plan which
each user modifies. Limitations of the fingerd program include: a)
One must already know which host to finger to find a specific person,
b) since primarily UNIX machines run fingerd, people who reside on
other types of operating systems are not locateable by this method,
c) fingerd is often disabled on UNIX systems for security purposes,
d) if one wishes to be found on more than one system, one must make
sure that all the .plan files are consistent, and e) there is no way
to search the .plan files on a given host to (for example) find
everyone on mazatzal.merit.edu who works on X.500. Thus, fingerd has
a limited usefulness as a piece of the Internet Directory.
The whois utility, which is available on a wide of variety of
systems, works by querying a centralized database maintained at the
DDN NIC, which was for many years located at SRI International in
Menlo Park, California, and is now located at GSI. This database
contains a large amount of information which primarily deals with
people and equipment which is used to build the Internet. SRI (and
now GSI) has been able to collect the information in the WHOIS
database as part of its role as the Network Information Center for
the TCP/IP portion of the Internet.
The whois utility is ubiquitous, and has a very simple interface. A
DISI Working Group [Page 3]
RFC 1309 Technical Overview of X.500 March 1992
typical whois query look like:
whois Reynolds
and returns information like:
Reynolds, John F. (JFR22) 532JFR@DOM1.NWAC.SEA06.NAVY.MIL
(702) 426-2604 (DSN) 830-2604
Reynolds, John J. (JJR40) amsel-lg-pl-a@MONMOUTH-EMH3.ARMY.MIL
(908) 532-3817 (DSN) 992-3817
Reynolds, John W. (JWR46) EAAV-AP@SEOUL-EMH1.ARMY.MIL
(DSN) 723-3358
Reynolds, Joseph T. (JTR10) JREYNOLDS@PAXRV-NES.NAVY.MIL
011-63-47-885-3194 (DSN) 885-3194
Reynolds, Joyce K. (JKR1) JKREY@ISI.EDU (213) 822-1511
Reynolds, Keith (KR35) keithr@SCO.CO (408) 425-7222
Reynolds, Kenneth (KR94) (502) 454-2950
Reynolds, Kevin A. (KR39) REYNOLDS@DUGWAY-EMH1.ARMY.MIL
(801) 831-5441 (DSN) 789-5441
Reynolds, Lee B. (LBR9) reynolds@TECHNET.NM.ORG (505) 345-6555
a further lookup on Joyce Reynolds with this command line:
whois JKR1
returns:
Reynolds, Joyce K. (JKR1) JKREY@ISI.EDU
University of Southern California
Information Sciences Institute
4676 Admiralty Way
Marina del Rey, CA 90292
(310) 822-1511
Record last updated on 07-Jan-91.
The whois database also contains information about Domain Name System
(DNS) and has some information about hosts, major regional networks,
and large parts of the MILNET system.
The WHOIS database is large enough and comprehensive enough to
exhibit many of the flaws of a large centralized database: a) As the
database is maintained on one machine, a processor bottleneck forces
slow response during times of peak querying activity, even if many of
these queries are unrelated, b) as the database is maintained on one
machine, a storage bottleneck forces the database administrators to
severely limit the amount of information which can be kept on each
entry in the database, c) all changes to the database have to be
DISI Working Group [Page 4]
RFC 1309 Technical Overview of X.500 March 1992
mailed to a "hostmaster" and then physically reentered into the
database, increasing both the turnaround time and the likelihood for
a mistake in transcription.
The Domain Name System is used in the Internet to keep track of host
to IP address mapping. The basic mechanism is that each domain, such
as merit.edu or k-12.edu, is registered with the NIC, and at time of
registration, a primary and (perhaps) some secondary nameservers are
identified for that domain. Each of these nameservers must provide
host name to IP address mapping for each host in the domain. Thus,
the nameservice is supplied in a distributed fashion. It is also
possible to split a domain into subdomains, with a different
nameserver for each subdomain.
Although in many cases one uses the DNS without being aware of it,
because humans prefer to remember names and not IP addresses, it is
possible to interactively query the DNS with the nslookup utility. A
sample session using the nslookup utility:
home.merit.edu(1): nslookup
Default Server: merit.edu
Address: 35.42.1.42
> scanf.merit.edu
Server: merit.edu
Address: 35.42.1.42
Name: scanf.merit.edu
Address: 35.42.1.92
> 35.42.1.92
Server: merit.edu
Address: 35.42.1.42
Name: [35.42.1.92]
Address: 35.42.1.92
Thus, we can explicitly determine the address associated with a given
host. Reverse name mapping is also possible with the DNS, as in this
example:
DISI Working Group [Page 5]
RFC 1309 Technical Overview of X.500 March 1992
home.merit.edu(2): traceroute ans.net
traceroute to ans.net (147.225.1.2), 30 hops max, 40 byte packets
1 t3peer (35.1.1.33) 11 ms 5 ms 5 ms
2 enss (35.1.1.1) 6 ms 6 ms 6 ms
.................
9 192.77.154.1 (192.77.154.1) 51 ms 43 ms 49 ms
10 nis.ans.net (147.225.1.2) 53 ms 53 ms 46 ms
At each hop of the traceroute, the program attempts to do a reverse
lookup through the DNS and displays the results when successful.
Although the DNS has served superlatively for the purpose it was
developed, i.e. to allow maintenance of the namespace in a
distributed fashion, and to provide very rapid lookups in the
namespace, there are, of course, some limitations. Although there has
been some discussion of including other types of information in the
DNS, to find a given person at this time, assuming you know where she
works, you have to use a combination of the DNS and finger to even
make a stab at finding her. Also, the DNS has very limited search
capabilities right now. The lack of search capabilities alone shows
that we cannot provide a rich Directory Service through the DNS.
3: THE X.500 MODEL OF DIRECTORY SERVICE
X.500 is a CCITT protocol which is designed to build a distributed,
global directory. It offers the following features:
* Decentralized Maintenance:
Each site running X.500 is responsible ONLY for its local part
of the Directory, so updates and maintenance can be done instantly.
* Powerful Searching Capabilities:
X.500 provides powerful searching facilities that allow users to
construct arbitrarily complex queries.
* Single Global Namespace:
Much like the DNS, X.500 provides a single homogeneous namespace
to users. The X.500 namespace is more flexible and expandable
than the DNS.
* Structured Information Framework:
X.500 defines the information framework used in the directory,
allowing local extensions.
DISI Working Group [Page 6]
RFC 1309 Technical Overview of X.500 March 1992
* Standards-Based Directory:
As X.500 can be used to build a standards-based directory,
applications which require directory information (e-mail,
automated resource locators, special-purpose directory tools)
can access a planet's worth of information in a uniform manner,
no matter where they are based or currently running.
The '88 version of the X.500 standard talks about 3 models required
to build the X.500 Directory Service: the Directory Model, the
Information Model, and the Security Model. In this section, we will
provide a brief overview of the Directory and Information Models
sufficient to explain the vast functionality of X.500.
To illustrate the Information Model, we will first show how
information is held in the Directory, then we will show what types of
information can be held in the Directory, and then we will see how
the information is arranged so that we can retrieve the desired
pieces from the Directory.
The primary construct holding information in the Directory is the
"entry". Each Directory entry contains information about one object;
for example, a person, a computer network, or an organization. Each
entry is built from a collection of "attributes", each of which holds
a single piece of information about the object. Some attributes which
might be used to build an entry for a person would be "surname",
"telephonenumber", "postaladdress", etc. Each attribute has an
associated "attribute syntax", which describes the type of data that
attribute contains, for example, photo data, a time code, or a string
of letters and numbers. As an example, let's look at part of an entry
for a person.
Entry for John Smith contains:
attribute ---> surName= Smith <--- attribute value
|---> telephoneNumber= 999-9999 <--- attribute value
|---> title= Janitor <--- attribute value
...
The attribute syntax for the surName attribute would be
CaseIgnoreString, which would tell X.500 that surName could contain
any string, and case would not matter; the attribute syntax for the
telephoneNumber attribute would be TelephoneNumber, which would
DISI Working Group [Page 7]
RFC 1309 Technical Overview of X.500 March 1992
specify that telephoneNumber could contain a string composed of
digits, dashes, parenthesis, and a plus sign. The attribute syntax
for the title attribute would also be CaseIgnoreString. A good
analogy in database terms for what we've seen so far might be to
think of a Directory entry as a database record, an attribute as a
field in that record, and an attribute syntax as a field type
(decimal number, string) for a field in a record.
At this point in our description of the information model, we have no
way of knowing what type of object a given entry represents. X.500
uses the concept of an "object class" to specify that information,
and an attribute named "objectClass" which each entry contains to
specify to which object class(es) the entry belongs.
Each object class in X.500 has a definition which lists the set of
mandatory attributes, which must be present, and a set of optional
attributes, which may be present, in an entry of that class. An given
object class A may be a subclass of another class B, in which case
object class A inherits all the mandatory and optional attributes of
B in addition to its own.
The object classes in X.500 are arranged in a hierarchical manner
according to class inheritance; the following diagram shows a part of
the object class hierarchy.
DISI Working Group [Page 8]
RFC 1309 Technical Overview of X.500 March 1992
_____________
| | "top" has one mandatory
| top | attribute "objectClass",
|_____________| and nooptional attributes.
| | |
| | | every other object class is a
________________| | | subclass of "top"...
| | ...
______|________ _____|_______
| | | |"organization" inherits one
| country | | organization |mandatory attribute from
|_______________| |_______________|"top", "objectClass"; adds one
more mandatory attribute "O"
"country" inherits one (for organization), and has
mandatory attribute from "top", many optional attributes. Any
"objectClass", adds one more subclass of "organization"
mandatory attribute "c" (for would inherit all of the
country), and has two optional mandatory and optional
attributes, "description" and attributes from "organization"
"searchGuide". Any subclass of including the attribute which
"country" would inherit all of the "organization" inherited
mandatory and optional attributes from "top".
of the "country" class, including
the attribute which "country"
inherited from "top".
Figure 1.
One major benefit of the object class concept is that it is in many
cases very easy to create a new object class which is only a slight
modification or extension of a previous class. For example, if I have
already defined an object class for "person" which contains a
person's name, phone number, address, and fax number, I can easily
define an "Internet person" object class by defining "Internet
person" as a subclass of "person", with the additional optional
attribute of "e-mail address". Thus in my definition of the "Internet
Person" object class, all my "person" type attributes are inherited
from "person". There are other benefits which are beyond the scope of
this paper.
X.500 hierarchically organizes the namespace in the Directory
Information Base (DIB); recall that this hierarchical organization is
called the Directory Information Tree (DIT). Each entry in the DIB
occupies a certain location in the DIT. An entry which has no
children is called a leaf entry, an entry which has children is
called a non-leaf node. Each entry in the DIT contains one or more
DISI Working Group [Page 9]
RFC 1309 Technical Overview of X.500 March 1992
attributes which together comprise the Relative Distinguished Name
(RDN) of that entry, there is a "root" entry (which has no
attributes, a special case) which forms the base node of the DIT. The
Distinguished Name of a specific entry is the sequence of RDNs of the
entries on the path from the root entry to the entry in question. A
diagram here will help to clarify this:
Level of DIT Root RDN Distinguished Name
root * nothing { }
/ | \
country (other / | \
things at this / | \ c=us {c=us}
level) c=gb c=us c=ca
/ | \
/ | \
/ | \
organization o=SRI o=Merit o=DEC o=Merit {c=us, o=Merit}
(other things / | \
at this level) / | \
/ | \
Third level cn=Chris Weider cn=Chris Weider {c=us, o=Merit,
cn=Chris Weider}
Figure 2: Building a DN from RDNs (adapted from a
diagram in the X.500 (88) Blue Book)
Each entry in this tree contains more attributes than have been shown
here, but in each case only one attribute for each entry has been
used for that entry's RDN. As noted above, any entry in the tree
could use more than one attribute to build its RDN. X.500 also allows
the use of alias names, so that the entry {c=us, o=Merit, cn=Chris
Weider} could be also found through an alias entry such as {c=us,
o=SRI, ou=FOX Project, cn=Drone 1} which would point to the first
entry.
Now that we've seen what kinds of information can be kept in the
Directory, we should look at how the Directory stores this
information and how a Directory users accesses the information. There
are two components of this model: a Directory User Agent (DUA), which
accesses the Directory on behalf of a user, and the Directory System
Agent, which can be viewed as holding a particular subset of the DIB,
and can also provide an access point to the Directory for a DUA.
Now, the entire DIB is distributed through the world-wide collection
of DSAs which form the Directory, and the DSAs employ two techniques
DISI Working Group [Page 10]
RFC 1309 Technical Overview of X.500 March 1992
to allow this distribution to be transparent to the user, called
"chaining" and "referral". The details of these two techniques would
take up another page, so it suffices to say that to each user, it
appears that the entire global directory is on her desktop. (Of
course, if the information requested is on the other side of the
world, it may seem that the desktop directory is a bit slow for that
request...)
To describe the functionality of X.500, we will need to separate
three stages in the evolution of X.500: 1) the 1988 standard, 2)
X.500 as implemented in QUIPU, and 3) the (proposed) 1992 standard.
We will list some of the features described in the 1988 standard,
show how they were implemented in QUIPU, and discuss where the 1992
standard will take us. The QUIPU implementation was chosen because
a) it is widely used in the U.S. and European Directory Services
Pilot projects, and b) it works well. For a survey of other X.500
implementations and a catalogue of DUAs, see [Lang].
There are a number of advantages that the X.500 Directory accrues
simply by virtue of the fact that it is distributed, not limited to a
single machine. Among these are:
* An enormously large potential namespace.
Since the Directory is not limited to a single machine, many
hundreds of machines can be used to store Directory entries.
* The ability to allow local administration of local data.
An organization or group can run a local DSA to master their
information, facilitating much more accurate data throughout
the Directory.
The functionality built into the X.500(88) standard includes:
* Advanced searching capabilities.
The Directory supports arbitrarily complex searches at an
attribute level. As the object classes a specific entry
belongs to is maintained in the objectClass attribute, this
also allows Directory searches for specific types of objects.
Thus, one could search the c=US subtree for anyone with a last
name beginning with S, who also has either a fax number in the
(313) area code or an e-mail address ending in umich.edu.
This feature of X.500 also helps to provide the basic
functionality for a Yellow Pages service.
DISI Working Group [Page 11]
RFC 1309 Technical Overview of X.500 March 1992
* A uniform namespace with local extensibility.
The Directory provides a uniform namespace, but local
specialized directories can also be implemented. Locally
defined extensions can include new object classes, new
attributes, and new attribute types.
* Security issues.
The X.500 (88) standards define two types of security for
Directory data: Simple Authentication (which uses passwords),
and Strong Authentication (which uses cryptographic keys).
Simple authentication has been widely implemented, strong
authentication has been less widely implemented. Each of
these authentication techniques are invoked when a user or
process attempts a Directory operation through a DUA.
In addition to the global benefits of the X.500 standard, there are
many local benefits. One can use their local DSA for company or
campus wide directory services; for example, the University of
Michigan is providing all the campus directory services through
X.500. The DUAs are available for a wide range of platforms,
including X-Windows systems and Macintoshes.
Functionality beyond the X.500 (88) standard implemented by QUIPU
includes:
* Access control lists.
An access control list is a way to provide security for each
attribute of an entry. For example, each attribute in a given
entry can be permitted for detect, compare, read, and modify
permissions based on the reader's membership in various groups.
For example, one can specify that some information in a given
entry is public, some can be read only by members of the
organization, and some can only be modified by the owner of
the entry.
* Replication.
Replication provides a method whereby frequently accessed
information in a DSA other than the local one can be kept by
the local DSA on a "slave" basis, with updates of the "slave"
data provided automatically by QUIPU from the "master" data
residing on the foreign DSA. This provides alternate access
points to that data, and can make searches and retrievals
more rapid as there is much less overhead in the form or
network transport.
DISI Working Group [Page 12]
RFC 1309 Technical Overview of X.500 March 1992
As flexible and forward looking as X.500 is, it certainly was not
designed to solve everyone's needs for all time to come. X.500 is not
a general purpose database, nor is it a Data Base Management System
(DBMS). X.500 defines no standards for output formats, and it
certainly doesn't have a report generation capability. The technical
mechanisms are not yet in place for the Directory to contain
information about itself, thus new attributes and new attribute types
are rather slowly distributed (by hand).
Searches can be slow, for two reasons: a) searches across a widely
distributed portion of the namespace (c=US, for example) has a delay
which is partially caused by network transmission times, and can be
compounded by implementations that cache the partial search returns
until everyone has reported back, and b) some implementations are
slow at searching anyway, and this is very sensitive to such things
as processor speed and available swap space. Another implementation
"problem" is a tradeoff with security for the Directory: most
implementations have an administrative limit on the amount of
information which can be returned for a specific search. For
example, if a search returns 1000 hits, 20 of those might be
displayed, with the rest lost. Thus a person performing a large
search might have to perform a number of small searches. This was
implemented because an organization might want to make it hard to
"troll" for the organization's entire database.
Also, there is at the moment no clear consensus on the ideal shape of
the DIT, or on the idea structure of the object tree. This can make
it hard to add to the current corpus of X.500 work, and the number of
RFCs on various aspects of the X.500 deployment is growing monthly.
Despite this, however, X.500 is very good at what it was designed to
do; i.e., to provide primary directory services and "resource
location" for a wide band oftypes of information.
The 1988 version of the X.500 standard proved to be quite sufficient
to start building a Directory Service. However, many of the new
functions implemented in QUIPU were necessary if the Directory were
to function in a reasonable manner. X.500 (92) will include
formalized and standardized versions of those advances, including
* A formalized replication procedure.
* Enhanced searching capacities.
DISI Working Group [Page 13]
RFC 1309 Technical Overview of X.500 March 1992
* Formalization of access control mechanisms, including access
control lists.
Each of these will provide a richer Directory, but you don't have to
wait for them! You can become part of the Directory today!
4: WHAT X.500 CAN DO FOR YOU TODAY
X.500 is filling Directory Services needs in a large number of
countries. As a directory to locate people, it is provided in the
U.S. as the White Pages Pilot Project, run by PSI, and in Europe
under the PARADISE Project as a series of nation-wide pilots. It is
also being used by the FOX Project in the United States to provide
WHOIS services for people and networks, and to provide directories of
objects as disparate as NIC Profiles and a pilot K-12 Educators
directory. It is also being investigated for its ability to provide
resource location facilities and to provide source location for WAIS
servers. In fact, in almost every area where one could imagine
needing a directory service (particularly for distributed directory
services), X.500 is either providing those services or being expanded
to provide those services.
In particular, X.500 was envisioned by its creators as providing
directory services for electronic mail, specifically for X.400. It is
being used in this fashion today at the University of Michigan:
everyone at the University has a unified mail address, e.g.
Chris.Weider@umich.edu. An X.500 server then reroutes that mail to
the appropriate user's real mail address in a transparent fashion.
Similarly, Sprint is using X.500 to administrate the address space
for its internal X.400 mail systems.
Those of us working on X.500 feel that X.500's strengths lie in
providing directory services for people and objects, and for
providing primary resource location for a large number of online
services. We think that X.500 is a major component (though not the
only one) of a global Yellow Pages service. We would also like to
encourage each of you to join your national pilot projects; the more
coverage we can get, the easier you will be able to find the people
you need to contact.
DISI Working Group [Page 14]
RFC 1309 Technical Overview of X.500 March 1992
For further information, the authors recommend the following
documents:
Weider, C., and J. Reynolds, "Executive Introduction to Directory
Services Using the X.500 Protocol", FYI 13, RFC 1308, ANS, ISI,
March 1992.
Lang, R., and R. Wright, Editors, "A Catalog of Available X.500
Implementations", FYI 11, RFC 1292, SRI International, Lawrence
Berkeley Laboratory, January 1992.
Barker, P., and S. Hardcastle-Kille, "The COSINE and Internet
X.500 Schema", RFC 1274, University College London, November 1991.
Hardcastle-Kille, S., "Replication Requirements to provide an
Internet Directory using X.500", RFC 1275, University College
London, November, 1991.
Hardcastle-Kille, S., "Replication and Distributed Operations
extensions to provide an Internet Directory using X.500", RFC
1276, University College London, November 1991.
Hardcastle-Kille, S., "Encoding Network Addresses to support
operation over non-OSI lower layers", RFC 1277, University College
London, November 1991.
Hardcastle-Kille, S., " A string encoding of Presentation
Address", RFC 1278, University College London, November 1991.
Hardcastle-Kille, S., "X.500 and Domains", RFC 1279, University
College London, November 1991.
Chris Weider
Advanced Network and Services, Inc.
2901 Hubbard G-1
Ann Arbor, MI 48105-2437
Phone (313) 663-2482
E-mail: weider@ans.net
Joyce K. Reynolds
Information Sciences Institute
University of Southern California
4676 Admirality Way
Marina del Rey, CA 90292
Phone: (310) 822-1511
EMail: jkrey@isi.edu
Sergio Heker
JvNCnet
Princeton University
6 von Neumann Hall
Princeton, NJ 08544
Phone: (609) 258-2400
Email: heker@nisc.jvnc.net
DISI Working Group [Page 16]