Integrating the directories of various operating systems isn't a simple matter. Michael lays out the challenge.
Meta-Directory Mindset
Integrating the directories of various operating systems isn't a simple matter. Michael lays out the challenge.
Yes, it’s true. Active Directory isn’t the only directory
service on the block. In fact, I’ve read estimates that
each Fortune 1000 company has over a hundred data stores
of information that could be considered a directory, if
not a directory service. Even small organizations will
run across many situations where user information is stored
in other application-specific data stores such as accounting
and payroll systems. And even though the goal is to have
a suite of directory-enabled applications, this idea is
going to take time. Some applications your business lifeblood
is running on don’t even have plans to follow the DEN
(Directory Enabled Networks) route.
This leaves the obvious solution—a tried and true one
that we know and love, called integration. Only this time
it’s directory service integration. It seems the further
we travel down the road, the closer we remain to home.
The Mighty Meta-Directory
Directory service integration has a more formal name,
coined by the Burton Group (www.tbg.com),
called meta-directories, which you may have heard mentioned
during the last few “years of the directory.” Labels aside,
a meta-directory is nothing more than an architectural
concept that covers the issues necessary to implement
an umbrella directory over other directories.
As with most concepts, vendors claim varying interpretations
of the meta-directory idea. They’re all mostly true, although
a central consideration is that a directory becomes a
meta-directory when it’s used as the focal point to manage
other directories. Not just any data store can be considered
a meta-directory. It must have at least the capability
to obtain information from other data stores through standard
or proprietary means, and then display that information
to users and applications through its own interface. Beyond
this characteristic, a meta-directory is largely an implementation
issue rather than a product issue.
As a result, this plants the subject of meta-directories
squarely into the political realm of information systems;
the primary stakes are data ownership and control of that
ownership.
Synchronization vs. Brokering
From a technical perspective there are two main approaches,
with a few variations, to organizing and maintaining the
information in meta-directories: synchronization of data
stores, and chaining—or brokering—of the data stores.
In most situations, they’ll probably both be used in combination
to provide a logically unified directory service.
Synchronization is done through the replication of information
from one data store to another. This is a necessary feature
of an independent directory service, such as NDS or AD,
in order to maintain accurate and authoritative information
within the distributed databases that make up the respective
directories. However, for the meta-directory, synchronization-specific
information in each independent directory must be replicated
to the other. The direction of this replication depends
on a political decision: which directory is the authoritative
master and which is the slave? In many cases, a directory
can be the master of some information areas and a slave
to other areas. In such cases, a directory holds a peer
status between participating directories. The key to the
synchronization concept is this: there are changes made
to the data store within at least one of the directories,
if not multiple directories.
Chaining, which is also referred to as brokering, is
the retrieval and display of information from another
data store without actually updating the local data store.
It’s essentially a data request. This is useful because
you lose the replication latency inherent in synchronous
replication models. Each time you request information,
it’s retrieved from the authoritative source without having
to worry about the replication schedule between two or
more different directory data stores. However, there’s
another side to brokering. You have to make sure the master
data source is always available in order to obtain valid
information (or even any information at all).
The X.500 Model
One good way to understand meta-directories is to look
at the original distributed “meta-directory” design, X.500,
and to use those terms generically to describe the similar
architectural components found in various specific products.
You can then discover how a particular meta-directory
product is designed and determine if it fits in with your
network design and infrastructure.
At the core of the X.500 model is an administratively
distributed database, which contains useful information
about an object, such as its characteristics and location
on the network. The X.500 term for the entire distributed
database is the directory information base (DIB); its
distribution components are directory system agents (DSA).
The DSA is a data store with a hierarchical structure
called the directory information tree (DIT). Each entry
in the DIT structure consists of one or more nodes, called
DSA-specific entries (DSE). A DSE with no subsequent entries,
or child entries, is called a leaf entry, and a DSE with
a child is called a non-leaf node.
The most important feature of the X.500 standard as it
relates to meta-directories is the DSA-distributed data
stores. Each database is managed by the organization that
is most concerned with the information in its respective
data store. This same concept, applied within an organization,
is the fundamental idea behind meta-directories. The meta-directory
distribution is accomplished via the various DIBs through
the different vendor DSA components. Each DSA contains
the unique portion of the DIT it’s responsible for maintaining.
Since the meta-directory concept involves widely distributed
data stores, it would be impractical to have the entire
database reside on one computer. Also, there’s a greater
chance of accurate data flowing throughout the entire
directory when the local owners of a DSA maintain the
information.
The LDAP Connection
Instead of X.500, however, Lightweight Directory Access
Protocol (LDAP), a subset of X.500, is proving to be the
Rosetta stone between proprietary directory data stores.
Ironically, X.500’s greatest role may ultimately be that
it was responsible for LDAP’s implementation and character.
LDAP solves several X.500 limitations that have impeded
widespread implementation. First, LDAP doesn’t require
synchronous communication between servers or clients.
Requests and responses may be exchanged in any order so
long as every request receives a response if required.
Another popular LDAP feature is the fact that it’s implemented
over TCP instead of OSI for communications between both
clients and servers.
LDAP is primarily considered an access protocol. That’s
because it was first developed as an alternative to DAP
as an entry point into X.500 directories. However, it
has grown to encompass a complete directory service and
is now both an access protocol and a distributed directory
service standard. Major directory service vendors with
X.500 characteristics, such as AD, NDS, and Domino, are
using LDAP as the core method by which their DIBs are
queried.
LDAP isn’t a complete rework of X.500. Both support a
hierarchical namespace using entries with object class
attributes. You needn’t choose between LDAP or X.500,
since LDAP and X.500 servers will inter-operate with LDAP
servers passing queries to X.500 DSAs; the results will
be returned to LDAP clients.
While LDAP APIs are promising and rapidly growing in
acceptance and support, it’s doubtful that all the application
developers relevant to your information system have added
LDAP APIs. Therefore, a third-tier architecture layer
may be necessary for immediate meta-directory implementation.
A component may be needed that can communicate with a
proprietary data store underneath and expose an LDAP interface
on top to pass information back and forth between the
meta-directory and the proprietary data store. Of course,
a given product must have enough critical mass in the
market to motivate the vendor to develop a connector.
This type of layered architecture is heavily dependent
on a solid physical network infrastructure focused on
availability. This shouldn’t be minimized. It’s critical
that your directory team work in tandem with your infrastructure
and application development teams to make sure each aspect
of the network is designed to interoperate efficiently.
The Power of Good Design
Another interesting common communication method will
be Directory Services Markup Language (DSML), an extension
and use of the XML standard. The DSML standard (www.dsml.org)
was initiated by Bowstreet (www.bow street.com) and is
now jointly promoted by Microsoft, Novell, Sun, IBM, and
others to further the standardized integration of directory
services. The DSML standard will be particularly useful
in distributing and sharing directory information across
company boundaries. It will also play an important part
in the development of e-commerce.
A well-designed meta-directory will have a degree of
flexibility, so that the control and management of the
information can be in various areas of the directory.
For example, the IS department should be allowed to manage
the groups and to access control lists for resources,
while the HR department should have control over personal
addresses and contact information. When HR flags someone
as terminated in the directory, this should flow down
to that person’s access throughout the system.
In addition, there will always be the need for local
control of information that shouldn’t be replicated out
to other data stores, but that may be retrieved with the
proper credentials through chaining.
In addition, it’s common for data stored locally to be
considered more important and therefore better cared for
than data in a centralized source. This means a well-designed
meta-directory will be designed with a centralized architecture,
but with pockets of local control to allow the care of
information on a daily basis.
To build a unifying data store from proprietary data
stores, it’s important for your meta-directory to have
a flexible namespace. That namespace should allow you
to instantiate common objects such as username with different
values from the different data stores. A messaging system
will most likely have a different username than an ERP
system or NOS name object. The meta-directory must map,
or join, these values using a common unique attribute
associated with the value; that attribute must be the
same across the various data stores. Common examples would
be unique values such as social security numbers or, for
the privacy-conscious, employee numbers.
These joins can be a combination of synchronized data
brought into the meta-directory store, and brokered information
in which the details are left in the original data store.
Regardless, this unification allows the storage and flexibility
in gathering security information such as private/public
keys, password lists for different applications, and other
authentication information. That security information
can then be transparently forwarded for the user to the
appropriate application as it’s accessed.
Is A Meta-Directory Even Possible?
Some people question whether adding yet another directory
solves the problem of too many directories. One answer
comes from Microsoft, with its acquisition of Zoomit and
that company’s VIA meta-directory product. Microsoft wants
to add more functionality to AD services so it can be
considered for meta-directory applications. However, I’d
question if any directory exists that could have the structure
or schema to support every type of information necessary
for every type of application.
Many issues remain regarding the implementation of a
widespread directory within a company or among companies.
For example, the underlying schemas of each directory
must map to each other, and the security context for each
directory or organizational unit must be managed to properly
flow downstream. While I think the concept of meta-directories
is here to stay, I also think we’re embarking on a much
more complicated journey than we ever anticipated.