OSF Distributed Computing Environment Frequently-asked questions

Skip to first unread message

Jon Mauney

May 20, 1998, 3:00:00 AM5/20/98

Posted-By: auto-faq
Archive-name: dce/faq
Revision: 1.35 1998/05/20 11:49:08
Posting-frequency: monthly

DCE Frequently Asked Questions

This document answers some Frequently Asked Questions concerning
the OSF Distributed Computing Environment.
It is posted monthly to comp.soft-sys.dce, as well as
comp.answers, and news.answers.

The questions (and answers) in this FAQ have been divided
into 3 tiers, for convenience, scalability, and because its
just the cool thing to do.

(*) Tier 1, User support, contains general questions about DCE
(*) Tier 2, Application support, contains technical questions about DCE
(*) Tier 3, Data Management, contains pointers to further information
about DCE

Tier 0 -- Header files

The document is maintained by Jon Mauney, j...@mauney.com

Suggestions, corrections, and updates are actively sought.

This FAQ is available for browsing at the Open Group's WWW server.

This FAQ is posted on Usenet to comp.answers and comp.soft-sys.dce
and is available for anonymous ftp from sites that archive those
newsgroups, such as rtfm.mit.edu and fine mirror sites worldwide.

Last modified on April 22, 1998

Changes since it was last posted on March 12, 1998

Q 1.03 : updated info on NT, OS/390

Q 2cc-05,3.01,3.07: updated the URLs for Transarc's web site

Q 2r-07, 2p-06, 2s-01, 2r-14, 2cc-02: Added URLs for examples at
Transarc's web site.

Added Q2p-07: Should I choose UDP or TCP for my RPC?

Changes since it was last posted on January 17, 1998

3.01 Add link to IBM's AIX DCE web page

Changes since it was last posted on November 6, 1997

2p02 Add information on IBM enhancements to security server to permit
large number of DCE principals to be created.

2cc06 Add clarification that cross-cell delegation is possible in
restricted circumstances.

3.01 Add link to Keys Botzum's DCE&Encina question and answer page

3.05 Add links to Linux DCE web pages by Andrew Sandoval and by Jim Doyle.

Tier 1 -- Front-end, user support

Q 1.01: What is DCE?
Q 1.02: What are the advantages of DCE?
Q 1.03: What platforms support DCE?
Q 1.04: What products use DCE?
Q 1.05: Is DCE an official standard?
Q 1.06: The DCE threads uses draft 4 of the standard, but
Posix threads standard has moved beyond draft 4.
Will DCE change to the most recent standard?
Q 1.07: Is anyone using DCE for real-life mission-critical systems?
Q 1.08: What is the relationship between DCE and CORBA?
Q 1.09: Is DCE IDL the same as all the other IDL's in the world?
Q 1.10: Now that OSF is the Open Group, should we talk about OG DCE?

Tier 2 -- Application support

Cell Configuration

Sub-interface: RPC and IDL
Q 2r-01: Several of the other questions concern "interfaces".
What is meant by an interface in DCE RPC?
Q 2r-02: Can a DCE client import multiple interfaces?
Q 2r-03: Can a DCE client connect to multiple servers?
Q 2r-04: Can a DCE server export multiple interfaces?
Q 2r-05: Can a process be both a server and a client?
Q 2r-06: How do I perform asynchronous RPC?
Q 2r-07: How can a server keep track of multiple clients?
Q 2r-08: How can a server detect that a client as exited or crashed?
Q 2r-09: How can a server identify the client that has called it?
Q 2r-10: Can I move idl-compiled stubs from one platform to another and
rebuild the object files locally?
Q 2r-11: How are UUIDs generated?
Q 2r-12: How can I pass a UUID as a parameter in an RPC?
What is the format of a uuid_t structure?
Q 2r-13: How can I pass a binding handle as an ordinary parameter in an RPC?
Q 2r-14: How do I keep my server from
advertising all the host machine's network addresses with CDS?
Q2r-15: How do I share type declarations between two
or more IDL files?
Q 2r-16: How can I control the timeout of a context handle?
Q 2r-17: A server has moved or otherwise changed its binding information.
How do I make the client obtain the new binding ?

Sub-Interface: Performance
Q 2p-01: How efficient is DCE RPC?
Q 2p-02: What is the practical limit on the size of a DCE cell?
Q 2p-03: How much memory and disk space is required for DCE services?
Q 2p-04: How can I control the number of concurrent client connections
and the size of the request queue on a server?
Q 2p-05: My server gets a stack error when sending large objects. How
can I avoid this?
Q 2p-06: How do I get clients to connect to the nearest server?
Q 2p-07: Should I choose UDP or TCP for my RPC?

Sub-Interface: Compatibility
Q 2c-01: Will Windows NT communicate with DCE?
Q 2c-02: Can I use DCE from C++?
Q 2c-03: Can I write an application that uses DCE and X11/Motif?
Q 2c-04: Is DCE RPC compatible with ONC RPC?
Q 2c-05: Is XDR compatible with NDR?

Sub-Interface: Cell Configuration
Q 2cc-01: Is it possible for a machine to be a member of more than one DCE cell?
Q 2cc-02: How do I configure cells to find each other using DNS?
Q 2cc-03: Is it possible for a user in one cell to use secure services
in another cell?
Q 2cc-04:: How do I change the name of my cell?
Q 2cc-05:: How do I change the IP address of
a host in my cell?
Q 2cc-06:: How do I do inter-cell delegation?
Q 2cc-07:: How can I find out who is currently logged in to a DCE cell?

Sub-Interface: Security
Q 2s-01: Where can I find an ACL manager to include in my application?
Q 2s-02: Does DCE Security interoperate with other Kerberos systems?
Q 2s-03: Can I allow DCE clients and servers to communicate
across a firewall?
Can I control the endpoints assigned to DCE servers?
Q 2s-04: Do DCE servers automatically update their long term secret keys?
Q 2s-05: What problems might I cause by changing the expiration time
or password lifespan in a running cell?
Q2s-06: How do I begin to force password changes,
without getting into the trouble described in the previous question?
Q 2s_07: Do all versions of DCE support the sec_key_mgmt_manage_key()
functionality in the same way?
Q 2s_08: How can I work around the key_mgmt problem if I have older client
machines, but DCE 1.1+ based servers?

Tier 3 -- Data management

Q 3.01: Where can I get online information about DCE?
Q 3.02: What books are published on DCE?
Q 3.03: Where can I get more information about DCE and DCE-related products?
Q 3.04: What are DCE RFCs, and how can I get them?
Q 3.05: Where can I get the Public Domain version of DCE?
Q 3.06: Is there a DCE Users Group I can join?
Q 3.07: Are there any example programs available on-line?

The answers please:
Tier 1 -- Front-end User Support

Q 1.01: What is DCE?

DCE is the Distributed Computing Environment, from the
Open Software Foundation. (It is called "the DCE" by sticklers
for grammatical consistency.) (The Open Software Foundation
is now called the Open Group.)

DCE consists of multiple components which have been integrated
to work closely together. They are the Remote Procedure Call
(RPC), the Cell and Global Directory Services (CDS and GDS), the
Security Service, DCE Threads, Distributed Time Service
(DTS),and Distributed File Service (DFS). The Threads, RPC, CDS,
Security, and DTS components are commonly referred to as the
"secure core" and are the required components of any DCE installation.
DFS is an optional component. DCE also includes administration tools
to manage these components.

DCE is called "middleware" or "enabling technology." It is not
intended to exist alone, but instead should be bundled into a
vendor's operating system offering, or integrated in by a third-party
vendor. DCE's security and distributed filesystem, for example,
can completely replace their current, non-network, analogs.
DCE is not an application in itself, but is used to build custom
applications or to support purchased applications.

Q 1.02: What are the advantages of DCE?


First, DCE provides services that can be found in other computer
networking environments, but packages them so as to make them
much easier to use. For example, the DCE Remote Procedure Call
(RPC) facility provides a way of communicating between software
modules running on different systems that is much simpler to code
than older methods, such as using socket calls.

Second, DCE provides new capabilities that go beyond what was
available previously. The DCE Security Service provides a
reliable way of determining if a user of a distributed system
should be allowed to perform a certain action, for example. This
is very useful for most distributed applications, yet the design
and implementation effort entailed in providing such a capability
would be prohibitive for an individual developer.

Third, DCE integrates components in a manner that makes them more
valuable together than separately. For example, the DCE RPC uses
threads in such a way that a developer can implement a
multi-threaded server without ever explicitly creating or
destroying a thread.

Finally, DCE supports both portability and interoperability by
providing the developer with capabilities that hide differences
among the various hardware, software and networking elements an
application will deal with in a large network. For example, the
RPC automatically converts data from the format used by one
computer to that used by another.

Portability is a measure of the ease with which a piece of
software that executes on one type of computer can be made to
execute on a different type of computer. Interoperability is a
measure of the ability of computers of different types to
participate in the same distributed system.
Q 1.03: What platforms support DCE?


DCE is fully supported on most major platforms, including all major
Unix platforms and many non-Unix platforms.
Most vendors support at least the "secure core" which means all of the
DCE services except the Distributed File Service and X.500 interface
to the Global Directory Service.

Some products are client-only, which means that the actual servers
for the DCE services are not provided: Directory Service, Security
Service, Time Service. Client machines can use these services,
they simply cannot run the server programs; another machine
in the cell must run the server programs. Application programs
can be built and run the application servers on these "client-only"

The status of DCE implementations changes frequently; check with the
specific vendors for the latest details, or see the DCE product
list (see the next question). Updates to this FAQ
are solicited.
This table summarizes the DCE implementations currently available:

Platform Comments
Unix platforms:
Digital UNIX: Secure core, DFS, and X.500
Ultrix had a "Starter kit", but of course Ultrix is
no longer supported...
HP-UX: Secure Core, DFS , and X.500
DG/UX Secure Core and X.500
AIX: Secure core, DFS, and X.500
SunOS, Secure Core
Solaris: Secure core and DFS
Implementation by Transarc, also re-sold by Sun
Irix Secure Core, DFS, X.500
SCO Secure Core was expected by August 1994
AT&T GIS, SVR4 on Intel Secure Core
Sinix (Mips and Intel) Secure Core, DFS, and X.500
Cray Unicos Secure Core Client, and DFS server
Pyramid DC/OSx (SVR4) Secure Core
Hitachi HI-OSF/1-MJ (osf/1) Secure Core, with Japanese on-line doc
Hitachi HI-UX/WE2 Secure Core
Sony NEWS (SVR4) Secure Core
Fujitsu DS/90 (SVR4) Secure Core
System V on Intel: Secure Core, from Gradient Technologies
Unixware, Sequent Dynix, NCR MP-RAS, Unisys
Dascom AD V1.1 (OSF/1 on Intel) Secure Core Client

Non-Unix platforms:
Tandem NonStop Kernel Secure core
OpenVms (Vax, and Alpha AXP) Secure core, X.500 support
Microsoft Windows: Client-only, available from Gradient Technologies
Gradient's port is also sold by Siemens,
and IBM.
Unimetrics has announced DFS client products
for Windows
Transarc has a "DFS-Light" client for Windows
Macintosh Client only, from Gradient Technologies
Open Environment Corporation has a Macintosh
client that allows access to DCE servers by
way of an OEC server on another machine.
Windows NT: out-of-the-box, NT contains a subset of
DCE RPC. See Q 2c-01 for further discussion.
Gradient has full secure core DCE+DFS for NT
Digital has secure core
IBM has secure core
Transarc has DFS client and server and a
"DFS-Light" client for WinNT
Windows 95 Client only, from Gradient
Client only, from Digital
Transarc has a "DFS-Light" client for Win95
OS/2: Secure core
HP 3000 with MPE/iX Secure Core
VM Kapsch AG supports DCE client and the
security server on IBM's VM and VM/ESA
VM/ESA IBM has announced intent to support
"selected components" of DCE on VM/ESA.
These components boil down to secure client
without DTS
Kapsch AG supports DCE client and the
security server on IBM's VM and VM/ESA
IBM OS/390 (MVS) secure core + DFS
IBM AS/400 client only
BS2000 Client Only
Bull DPX/20 Secure Core, X.500, DFS
GCOS on Bull DPS 9xxx/7xxx threadless client only

Q 1.04: What products use DCE?


The Open Group maintains an Open Software Registry, which contains
information about DCE products, among others. Access is free, but
you must register.
The registry is available on-line at http://www.opensoftware.com/
Once you get in, you can search for DCE, and your favorite platform
or other criteria.
Hint: A search using "DCE" as the product name will find only
those products whose listed name starts with DCE. If you want
to find products like "PC-DCE", "Nonstop DCE", and "Kapsch-DCE/VM", search
for "*DCE" instead. If you want to find DCE-related products like Encina
or Dazel, go to the advanced search and choose the Keyword DCE.
Note to vendors, not all DCE implementations show up under a keyword
search for DCE, so check your listings.

Q 1.05: Is DCE an official standard?


The OSF calls the specification an Application Environment Specification,
or AES. The AES documents both the software programming interfaces and
also the communications protocols employed by DCE. Thus it would be
possible, in theory, for someone to build a compatible
implementation without using the code from the Open Group.
The AES for RPC, Time, and Directory services have been accepted as
standards by by X/Open. The AES for Security is currently undergoing

DCE Threads follow the Posix Threads draft standard 1003.4a draft
4. DCE Access Control Lists (ACLs) are based on POSIX.6 Draft
12. The Distributed Time Service (DTS) uses time formats defined
by international standards and in POSIX.4. The Global Directory
Service (GDS) complies with the X.500 international standard.
(Although DCE complies with the 1988 version of X.500, not
the 1992 version.)

ISO is considering an RPC standard based on the X/Open document.

DCE's status as a de facto standard is even stronger. Almost
every major hardware and software vendor has committed to
providing DCE on its platform. These vendors include not only
Open Group stalwarts such as IBM, DEC and HP, but also other key vendors
such as Novell, Inc. See Q 1.03 for a list of DCE vendors. In addition,
a number of major user organizations (e.g., the European Economic
Community) have already embraced DCE as their standard for
distributed applications.

Q 1.06: The DCE threads uses draft 4 of the standard, but
Posix threads standard has moved beyond draft 4.
Will DCE change to the most recent standard?


It is hard to predict exactly what will happen.
But Open Group prefers to follow standards rather then invent them.
Now that Posix has settled on the standard, we can expect
the DCE to migrate to it.

Q 1.07: Is anyone using DCE for real-life mission-critical systems?


Yes, and more every day. Tokio Marine and Fire Insurance,
Co., Lehman Brothers, and Charles Schwab have all publically
described their ongoing rollout of DCE based applications.

The Open Group's Web server has a section that includes some reports on
'real-world' experiences from companies using DCE in production.

Intellisoft published a 48-page "advutorial" (that
means advertisement and tutorial) with the help of
many DCE vendors, totally devoted to the above. There are storeis about
DCE in production, technical essays, etc.
The supplement is on-line and available via Intellisoft's web site
(http://www.isoft.com/publications.html -- click on "DCE and The Enterprise" ).

Transarc's web server includes a page in which they describe solutions
used by some of their customers.


Q 1.08: What is the relationship between DCE and CORBA?


The short answer:

There is not a lot of direct relationship. DCE and CORBA
are tools to help you build distributed systems. Each has
its advantages and disadvantages. Use of one will not
hinder future use of the other.

DCE provides a lower-level programming model than does CORBA.
DCE is not fully "Object-Oriented". DCE has far better
inter-operability than (current) CORBA products.
DCE is an optional interoperability mechanism in the
CORBA 2.0 specification.

The long answer:
In order to understand the relationship between DCE and the Common
Object Request Broker Architecture (CORBA) of the Object Management
Group (OMG), it is necessary to consider the past, the present and the

Past - Historically, the object paradigm has been viewed as a break
with procedural styles of the past. Objects, which encapsulate data
and procedures behind an external interface, are often contrasted with
other approaches where procedures and data are treated separately.

In this context, DCE is a descendant of the procedural school which
emphasizes the decomposition of programs into procedures and achieves
distribution by locating some of those procedures remotely. Thus there
was a tendency for the object community, including the OMG, to view DCE
as technology which was obsolete before it was available.

However this view ignored the fact that designers of distributed
systems had for a long time recognized that the most successful
approach to developing distributed systems was to created encapsulated
objects that can only be accessed via well defined interfaces. Thus
the cornerstone of DCE RPC is the interface definition language (IDL)
which allows the external attributes of a set of server operations to
be specified.

Furthermore, the name-based binding mechanisms of DCE were extended to
include the ability to bind to a server based on the object instances
which it supports. These object binding mechanisms also allow the
transparent selection among multiple implementations of the same server
operations based on the type of the specified object. In object
terminology this is called polymorphism.

The DCE notion of a server supporting interfaces consisting of one or
more operations is so close to the notion of an object which provides
one or more methods, that it should be no surprise that CORBA defines
an IDL which differs from DCE IDL in only a few significant respects.

Principal among these is that in CORBA IDL every call must specify an
object, which is used in determining the server to use. DCE can do
this as well, but there is more work involved and it is optional.
Another difference is that CORBA IDL allows an interface to be defined
as a extension of one or more other interfaces, this is called
interface inheritance. DCE does not permit interface inheritance, but
may in the future. Implementation inheritance is not specified by
either DCE or CORBA.

The use of object oriented techniques and principles should not be
confused with using an object oriented language. Object oriented
designs can be expressed in procedural languages, and in fact most of
the current object environments supported C before supporting C++ or
Smalltalk. Therefore, the fact that the DCE API is implemented in C is
no barrier to using it to create a distributed object system. In fact,
CORBA specified C language bindings first.

Present - CORBA should not be viewed in isolation, but in the context
of all of the OMG's standardization efforts. OMG has defined a
reference architecture (OMA) and has defined or is defining standards
in a broad range of areas, including: databases, events, lifecycle,
transactions, persistence, security, naming and relationships. Viewed
in this way, OMG's activities are much more ambitious and broader in
scope than DCE.

A recent addition to CORBA, as a part of the CORBA 2.0 work was the
definition of the means of interoperability between ORBs. CORBA 2.0
defines one mandatory and two optional mechanisms. The mandatory means
is a new, lightweight protocol called UNO. The optional means are 1)
via a gateway and 2) via an alternative protocol definition. At the
present the only alternative protocol that has been defined is DCE

Many people who had hoped that DCE would be selected as the mandatory
protocol were disappointed at this result. However, it should be
observed that DCE is endorsed as alternative protocol and that several
vendors have committed to providing ORBs that interoperate via DCE.

Another difference between DCE and the OMG standards is one of general
philosophy. DCE has been defined quite rigorously in a series of
documents published by X/Open. There is a set of conformance tests
that are available to anyone. Any product passing these rigorous tests
can be branded as DCE, without necessarily being based on the OpenGroup
code. Several vendors, including Microsoft and Tandem have reverse
engineered significant portions of DCE.

OMG standards vary considerably in their level of detail, but in
general, aim at a much looser level of standardization. In some cases,
the standard merely specifies an object interface and some general
semantics. This approach is a deliberate attempt to encourage diverse
solutions which may be applicable in different environments. Even
where specifications are relatively tight, for example in the area of
CORBA portability, there is still room for considerable interpretation,
as witness the fact that there is at least one company that provides
consulting services on how to make CORBA applications compliant in

At the present time, CORBA-compliant products and products that work
with them do not provide a scaleable infrastructure suitable for large
environments. Key features such as concurrency mechanisms, security
and distributed transactions are not currently available. In contrast,
DCE provides proven heterogeneous interoperability and most of the
capabilities required by robust, production applications. Additional
capabilities can be obtained by means of third products, such as
transaction monitors built upon DCE. This situation will change over
the next 2-3 years from a combination of standardization work by OMG
and new product development by vendors.

Future - Most authorities agree that in the long term object technology
will be the basis for building large scale distributed systems. In
addition to the principle of encapsulation, object- based systems allow
systems to be built up, evolve and be reconfigured as needed because of
their ability to dynamically bind requesters to objects that provide

There are many specific issues concerning the properties of distributed
object systems that are the subject of research and debate. It is also
clear that there are some features of existing local object
environments and languages that will not scale effectively to large
scale distributed systems -- dynamic inheritance is one. Never the
less, the general direction of the future is clear.

Clearly, the high level of interest in OMG defined standards comes not
from current products, but from their exciting future potential. There
is a natural tendency to compare DCE's current capabilities with the
promise of CORBA's future. However, DCE is also evolving and will
likely add additional object oriented features in the future. For
example, HP is offering a DCE C++ class library which is expected to
eventually become a standard part of DCE.

Where DCE was built by integrating existing software, OMG has chosen by
and large to start with a clean sheet of paper. The idea is to be
better able to implement object oriented constructs without the baggage
of features carried over from previous systems. However, OMG faces
great challenges. Object theory is currently in a great state of
flux. Experts disagree on very fundamental issues about what features
are necessary, useful or harmful.

Developing standards under these conditions is extremely challenging.
OMG's approach to date has been to compromise and allow multiple
alternatives. It is unclear whether this will succeed in the long

Does the conclusion that future distributed systems will be object-
based mean that it is a mistake to build distributed systems today
using DCE? The answer is no for several reasons. First, many
organizations cannot afford to do nothing for several years. End users
have pressing needs for robust, scaleable systems today. For many
organizations, waiting would mean attempting to catch up with
competitors who will have a tremendous head start.

Second, as this brief discussion has shown, it is possible to employ
object techniques when developing distributed applications using DCE.
Carefully designed systems will be able to take advantage DCE features
such as dynamic binding and polymorphism and converge with
CORBA-compliant systems as they mature.

Third, if object environments are to be successful in supporting
industrial-strength distributed systems, they will have to address the
problems that DCE addresses. The skills and techniques developed in
working with DCE will be directly applicable to distributed systems
environments of the future. This applies not only software developers,
but also to operations personnel, planners, even business managers.

Further, the likelihood that DCE will be at least one technology for
CORBA interoperability, implies that the eventually migration of
applications which use DCE directly to an object environment should not
present any insurmountable difficulties.

Finally, your direct experience in developing and operating robust
distributed systems will provide you with great insight into the
important characteristics of distributed systems environments as they
apply to your organization's applications. This knowledge is vital to
the shaping of successful tools of the future. History has shown that
vendors and standards bodies, left to their own devices, will often
miss the mark.

Q 1.09: Is DCE IDL the same as all the other IDL's in the world?



IDL stands for "Interface Definition Language," and the idea
of using a special language to define the interface between
entities is not unique to DCE. In particular, OMG's IDL for CORBA
is used for the same purpose as DCE's, but the two languages
are not identical; see Q 1.08 for more information. There are
other Interface Definition Languages as well.
IDL also stands for "Interactive Data Language", which
is a completely unrelated product.

When asking or answering a question about IDL, one should be
careful about specifying which IDL is involved.


Q 1.10: Now that OSF is the Open Group, should we talk about OG DCE?

No. "OSF DCE" is the registered trademark, and will apparently remain the
name of the product.

Tier 2 -- Application Support

Q 2r-01: Several of the other questions concern "interfaces".
What is meant by an interface in DCE RPC?


An interface is a set of remote procedure call operations and
associated data. Every interface contains one or more
operations. An operation is an actual remote procedure. Each
operation may have input and output parameters associated with
it, just like any procedure call.

Q 2r-02: Can a DCE client import multiple interfaces?


Yes. A client can use as many different services as it needs.
To code such a client, simply include the header files for
all the RPC interfaces used, and code each call the same way
you would if using that interface in isolation.

Q 2r-03: Can a DCE client connect to multiple servers?


Yes. A client can connect to multiple servers providing different
services, and/or multiple servers providing the same service.

To use multiple servers with the same interface, the client must
obtain a binding handle for each server and use explicit handles
in the RPC.

Q 2r-04: Can a DCE server export multiple interfaces?


Yes. A server can provide service on multiple interfaces
simultaneously. A common example is a server that exports an
application interface and a management interface.

To code such an application, repeat the calls to rpc_server_register_if(),
rpc_ep_register(), (and rpc_ns_binding_export() if you do that sort
of thing in your server) for each interface, before calling

Q 2r-05: Can a process be both a server and a client?


Yes. There are two scenarios.

1) A program might act as a server for interface A,
and also as a client for interface B. This is easy.
The program merely imports interface B like a normal
client and exports interface A like a normal server.
2) A program might want to provide a service, and also
act as a client to other servers that provide the same
service. In this case, the programmer must expend more
effort. The problem is that the names of the server-side
functions (manager routines) clash with the names of the
client stubs. The solution is to manually build an
entry point vector for the server, and use different names
for the manager routines. For details on using entry point
vectors, see the Lockhart book (Q 3.02).

Note that most server programs also act as clients, since they
usually access the endpoint mapper (rpcd), and the security
service; these actions use RPCs, though it may not
be obvious in the code.

Q 2r-06: How do I perform asynchronous RPC?


DCE-RPC is synchronous. The way to make an asynchronous call is
to create a thread for each RPC call. You should be able to have
dozens, if not hundreds, of threads with no problem.

Q 2r-07: How can a server keep track of multiple clients?
For example, to know what information has already been
provided to a client, and thus vary subsequent responses.


The DCE RPC mechanism includes a "context handle" which can be
created by a server and returned to a client. The handle is used
on subsequent RPCs to identify the client.

includes sample code for a greet-type client/server
that uses context handles.

Q 2r-08: How can a server detect that a client has exited or crashed?


The context handle provides this ability. When a context handle
is created and passed to the client, the DCE runtime library
keeps track of the connection between client and server;
this may be done in the network code as in the case of TCP,
or by DCE-specific ping messages if a connectionless protocol is used.
When the client dies, the server is notified and executes a "rundown"
function to clean up its data structures.

Q 2r-09: How can a server identify the client that has called it?


The details vary, depending on exactly what you want to
identify about the client, but it all comes down to handles.
The server must receive a binding handle in order receive
information about the client. If there is not a binding
handle in the parameter list as defined in the IDL file, then
use the [explicit_handle] attribute in the ACF file to add
one when building the server.

If you want to identify the host machine on which the client
is executing, simply convert the binding handle to a string
binding using rpc_binding_to_string_binding() and then
extract the address using rpc_string_binding_parse().
When the server receives the handle, it contains the client

If you want to identify the human user who is running the client program,
you can get that from the authentication information in the handle.
rpc_binding_inq_auth_client() will give you the caller's PAC,
which contains, among other things, the principal's uuid
and name.

If you want to identify the particular client process so your
server can keep data specific to each client, then you need
to look into context handles as described in Q 2r-07.

Q 2r-10: Can I move idl-compiled stubs from one platform to another and
rebuild the object files locally?


No. You must run the IDL compiler separately on each platform.

The IDL compiler builds the client and server stubs to handle
network communication and data marshalling, which are
platform-specific activities. Therefore the stub code is not
portable and must be re-created on each platform. Likewise, while
the task of the stub does not change, the set of service routines
called from the stub may be changed by the vendor for any given
platform. Therefore stubs for the same RPC may look very different
on different platforms.

Q 2r-11: How are UUIDs generated?


The DCE AES specifies the method for generating UUIDs, in case
you want the gory details. In short, the UUID is a combination
of the network address and current time at the moment and place
it was generated. Assuming that network addresses are unique
and that time never runs backwards, this guarantees that UUIDs
really are unique.

Some implementations of uuidgen use the hardware address of the
ethernet adapter, some use the IP address of the host.

It is possible to run uuidgen on machines without any network
connection. In this case, the host must be assigned an address.
The AES specifies a file that can hold the network address.

The code for generating UUIDs is part of the publically available
DCE source code. See Q 3.05: for information
on obtaining the code.

Q 2r-12: How can I pass a UUID as a parameter in an RPC?
What is the format of a uuid_t structure?


The uuid_t structure is defined in <dce/nbase.idl>
All you have to do is import this into your .idl file
and pass parameters of that type all day long.

Note that uuid's passed as parameters are just ordinary data.
If you want the uuid to affect the interface or object
in the rpc, they must be attached to the binding handle.

Q 2r-13: How can I pass a binding handle as an ordinary parameter in an RPC?


The binding handle is an opaque type and cannot be marshalled
as data. A parameter of type handle_t must be the first parameter
and will serve to specify the binding of the RPC; an attempt
to use a handle_t as other than the first parameter will be flagged
as an error by the IDL compiler.

To send a handle from one process to another, convert it to
a string binding and pass the string as a parameter.

Q 2r-14: How do I keep my server from
advertising all the host machine's network addresses with CDS?

If a server machine has two (or more) network interfaces, it will
have at least that many network addresses. A typical server program
will export binding handles to CDS for every protocol sequence
on every network interface. If some of these interfaces are unreachable
from some clients, binding may be delayed while the client cycles
through the useless handles.

To limit a particular server,
set the environment variables RPC_UNSUPPORTED_NETIFS and/or
RPC_UNSUPPORTED_NETADDRS to exclude the interface or specific addresses
that you don't want your server to use. On HP, set RPC_SUPPORTED_NETADDRS
instead, to give a list of addresses you do want to use. Look in your
vendor's documentation for syntax and to verify that they support this
feature; it is not standardized by Open Group.
You must set the environment before starting the server, of course,
and you may have to manually remove CDS entries that are already present.

Q2r-15: How do I share type declarations between two or more IDL files?

Imagine a program that uses two interfaces, in two idl files, and
these two interfaces share many of the same structures
which must be typedef'ed in the idl file. A C-programmer's first
instinct would be to #include a header file in both idl files, and
this would appear to work, because the IDL compile generally runs
the C preprocessor. But when the application code is compiled and
#includes both of the generated *.h files, the C compiler will
complain about duplicated declarations of these structures.

You should instead put the shared declarations in a separate .idl file,
compile it separately with the idl compiler, and "import"
the .idl file into all the other idl files that need it.

This way, each definition will appear in exactly 1 generated .h file,
where it will be protected by the usual #ifdef to prevent multiple

Some shops have an explicit style guideline that says an idl file
should contain either type definitions or function declarations,
but not both. You might want to try it.


Q 2r-16: How can I control the timeout of a context handle?
Sometimes, the context-rundown routine gets called immediately when the
client is killed, and sometimes it takes a (relatively) long time. Why?

The DCE runtime calls the context rundown routine as soon as it notices
that the connection with the client has been lost. Exactly when the runtime
will notice depends on the communications protocol.

If the conversation is using a connection-oriented protocol like TCP,
then the runtime depends on the protocol to notify it of a lost connection;
if the client crashes or exits, the speed with which the server is
notified depends on the TCP implementation and may be quick or very slow.
If the socket is shut down gracefully, the rundown may be called
immediately; otherwise it all depends on the timeout parameters in TCP.
You may be able to adjust the timing by changing parameters in your
network stack, but it will affect the whole system, not just DCE.

With a connectionless protocol like UDP, the runtime generates its own
keepalive messages to test the "connection". (Actually, they are
"I'm Not Dead Yet" messages from the client.) If the client exits, the
server will assume the connection is down after a number of missed

Thus the speed with which your server is notified after a client crashes
may vary from platform to platform due to differences in TCP implementation,
and from minute to minute on the same platform depending on whether the
client happens to use a TCP or UDP binding handle.

Moral: context rundown allows the server to clean up, but is not
a real-time monitoring tool.

Q 2r-17: A server has moved or otherwise changed its binding information.
How do I make the client obtain the new binding ?
First, remember that a client must always be prepared to deal with
stale binding handles. Before starting to make RPCs, call
rpc_mgmt_is_server_listening() to verify the handle, and continue
importing bindings until you get one that works.

If the server information has changed recently, the information in
the client's cache may need updating. Call rpc_ns_mgmt_set_exp_age()
with an expiration_age value of 0 to force an update.

Q 2p-01: How efficient is DCE RPC?


Performance testing at several user organizations has shown that
DCE RPC performance is similiar to other RPC implementations when
doing the same things. The throughput and response times for a
series of remote procedure calls is similiar.

The use of features in DCE not present in other implementations
may consume additional time and resources. For example,
name-based binding may required additional time, depending on the
number of directories traversed. Using the packet integrity and
packet privacy features of the security service can increase
processing times as a linear function of message sizes.

There are three papers providing preliminary performance data
published in:

DCE--The OSF Distributed Computing Environment,
Lecture Notes in Computer Science #731, Springer-Verlag,
(see Q 3.02 for ISBN)

IBM has done quite a bit of performance testing of DCE. Many
of the reports are available on line; go to IBM's corporate
Web page, http://www.ibm.com, and choose the search feature to
search for something like "dce performance". To summarize a few
of the results in these reports:

(*) RPC time increases linearly with the size of the data.
(*) Passing a large array in one RPC call is about the same speed
as using a pipe
(*) Passing a large array in several RPC calls is slower than
a single call using a pipe
(*) The "packet integrity" security level slows the RPC, nearly
doubling the total time for calls.
(*) The "packet privacy" level incurs a several-fold increase in time.
(*) RPC is slower than simply making socket calls directly.
(*) The time spent on RPC overhead was a small fraction of the total
processing time in a realistic business scenario.
(*) RPC is pretty fast on Intel procesors (running OS/2).

Q 2p-02: What is the practical limit on the size of a DCE cell?


Good question. There are no hard and fast answers, but there are
some large-ish cells in operation.

Certainly it is reasonable to plan on cells with thousands of
nodes and perhaps tens of thousands of users.

The University of Michigan Center for Information Technology
Integration has done a study in which they added 50,000 entries
to the Cell Directory and to the security registry. Their results
are reported in technical reports 93-12 and 94-1. See Q 3.01 for
the ftp site for CITI tech reports.

Lexis-Nexis has a more recent study in which they added 400,000 accounts
to the DCE registry. The write-up is available at

Several years ago
IBM's DCE performance testing (see Q 2p-01 above for citation)
includes load testing to simulate the environment of a large cell.
A short, oversimplified, summary is that server machines based
on 486/66 and P90 (running OS/2) could support the projected needs
of a cell with 10,000 users. An informal post on Usenet claimed that
a test on an AIX platform had reached over 700,000 users created in the
security server.

IBM's OS/390 DCE server includes modifications to the security server to
use a better backing store so that the entire set of principals need not
be keep in-core.

Q 2p-03: How much memory and disk space is required for DCE services?


This depends on the size of the cell, the number of users,
number of services, etc.

According to a paper present by Dan Hamel of Transarc, at the
Decorum conference in February 1994, the following can be used
as rough guidelines:

(*) Security server: 2k per principal/account
same at replicated sites
(*) Directory server: 10k per directory, 1k per object
same at replicated sites
(*) end-user machines: Each dce_login creates new credential files,
which can build up. Space usage can range
from less than 1k to over 100k.

The security server as shipped by the Open Group keeps all security registry
information resident in main memory, which means that the security
host machine needs memory as well as disk.

Q 2p-04: How can I control the number of concurrent client connections
and the size of the request queue on a server?


The rpc_server_listen() call creates a pool of threads to
handle incoming RPCs. The server can be actively working on
at most this many RPCs simultaneously. The size of the pool
is controlled by a parameter to rpc_server_listen. This is
the primary place where you are expected to control the number
of simultaneous connections.

If you want to go deeper, into the question of what happens
when all the server-listen threads are busy, things become
tricky, because there are so many factors affecting the answer.

If all the server threads are currently busy, incoming requests
will be queued by the RPC runtime. There is not an explicit
API for controlling the size of this queue. The rpc_use_protseq*()
calls have a "max calls" parameter, but its effect is murky and
it may be ignored completely by the implementation.

If the RPC queue is full, the runtime will not accept any more
network messages. Incoming requests will therefore be held at
the network protocol layer, which maintains a queue of its own.
If the network layer's queue is full then the incoming message will
be rejected and the client will be told that it is unable to connect
to the server.

Q 2p-05: My server gets a stack error when sending large objects. How
can I avoid this?


Each thread in a process is assigned a fixed area for its
procedure-call stack.
The stubs normally marshall and unmarshall parameters in space
allocated on the thread's stack. If the parameters are large,
the stack size may be exceeded. In most thread implementations,
the stack size cannot be increased after the thread is created.
For threads created explicitly by your application, you can adjust
the size of the thread stack by setting an attribute before calling
However, server threads are created automatically, so that method
won't work; instead, call rpc_mgmt_set_server_stack_size() before
starting the threads with rpc_server_listen().

Another possibility is to use the [heap] attribute to have
some parameter types marshalled on the heap instead of the stack.

You should know that current implementations of the IDL compiler
generate recursive code to marshall linked lists. Therefore, passing
a long linked list may cause stack overflow due to all the
recursive calls.

Q 2p-06: How do I get clients to connect to the nearest server?


It's not easy. Currently, vanilla DCE gives essentially no help.

For clients for which you have source code, you can use the
rpc_ns_binding_lookup*() routines to search through the binding
handles and apply some custom logic to finding a nearby server.

For standard DCE services, such as CDS and security queries,
you get a random server, which could be anywhere in the cell.
For the security service, you might be tempted to set the
BIND_PE_SITE environment variable and the file
/opt/dcelocal/etc/security/pe_site to hardwire each client host
to its nearest security replica, but this is a hack with serious
administrative downside: if the replica goes down or is moved,
all the associated clients must be manually updated.

Individual products may provide additional help. For example
Transarc's DCE now includes a feature that lets clients connect to the
closest CDS server. Details at

Q 2p-07: Should I choose UDP or TCP for my RPC?


First have you considered the advantages of using all protocol sequences?
RPCs work the same either way. DCE's implementation provides the same
security and reliability over all communication protocols. In most cases
the performance is about the same as well.

For some applications and some platforms, there may be reasons to prefer
one protocol over another. For instance, if your platform has a poor
implementation of UDP, that could be a reason to prefer TCP, or vice-versa.

For heavily-loaded servers, UDP is often recommended over TCP.
In low-bandwidth networks, or when running servers under a debugger,
TCP may be preferred.

Transarc has a web page providing more information on the relative
merits of UDP and TCP, at


Cell Configuration

Q 2cc-01: Is it possible for a machine to be a member of more than one DCE cell?


No. A machine can only be in a single cell. However, it is
possible for cells to cooperate. See the next question.

DCE v1.1 will allow for "hierarchical cells", which may
solve the problem, depending on why you want to have a machine
in two cells.

Q 2cc-02: How do I configure cells to find each other using DNS?


In order for a client program in cell A to find a server in cell B,
it must be able to find the CDS server in cell B; once communication
with CDS is established, everything else in the cell can be found.
One of the ways to make the location of of the cell known is to
create the proper entries in the Internet Domain Name Service (DNS).

Suppose we want to make a cell available, and that the name
of the cell is "/.../cell.mauney.com". We must create
two DNS entries. One entry will provide the IP address of
the machine that runs the CDS server within the cell.
The other entry gives other details about the cell.

The first entry, which advertises the location of the CDS server,
can be either an MX or an AFSDB record type. The DCE code
will be happy to use either one, but you as a system administrator
have to make a choice; both possibilities have drawbacks.
The MX record type is intended for mail exchange, so using it
for DCE is a bit of a kludge; and if you already
have a machine in your name with the same hostname as your cell
name and it ever participates in e-mail, then you'll have a conflict.
On the other hand, the AFSDB record is new and not supported by
all nameservers; you'll need Bind version 4.9.2 or later.

The MX or AFSDB record relates the cell name to a host name.
DCE ignores the "preference" value of an MX record. With AFSDB,
you should use sub-type 2. E.g.,

cell.mauney.com. IN MX 200 tophat


cell.mauney.com. IN AFSDB 2 tophat

where "tophat" is the hostname of the machine that runs the CDS master

The second piece of information in DNS is a TXT record giving
UUIDs for the CDS master replica and clearinghouse. You obtain
the information with the command "cdscp show cell as dns", and
copy the information verbatim into DNS. E.g.

cell.mauney.com. IN TXT "1
44007e75-08b4-11ce-9055-08002b32b23b Master /.../cell.mauney.com/tophat_ch 43373550-08b4-11ce-9055-08002b32b23b tophat.mauney.com"

for a description of how it works, and some troubleshooting
hints in case it doesn't.

Q 2cc-03: Is it possible for a user in one cell to use secure services
in another cell?


Yes. The Access Control List (ACL) permits three entry types --
foreign_user, foreign_group and foreign_other -- which specify the
permissions available to users on other cells. All that is
required for intercell access, other than physical connectivity,
is for the two cell's security services to be configured to know
about each other.

There is a command, the rgy_edit "cell" command, that must be run, once,
by the cell admin of the two cells that wish to communicate. After
that, it's all transparent.

Q2cc-04:: How do I change the name of my cell?

You don't.

The cell name in embedded into too many hidden places. Live with
the old cell name, or create a new cell and configure it to look
like the old cell.

If you haven't created your cell yet, learn this lesson and
choose a cell name you can live with.

Q2cc-05:: How do I change the IP address of
a host in my cell?

Very Carefully.
You need start this before changing the address of the machine.

Transarc has several files of tips on the subject.
See. http://www.transarc.com/Public/Support/dce/admin_examples/misc/ip.html.

Q2cc-05:: How do I do inter-cell delegation?


Our initial answer to that question was:
Inter-cell delegation is a poor idea from a security standpoint.
Therefore, DCE does not implement it. That answer may have been too
absolute. It may be better to say that DCE severely restricts delegation
because of the security problems.

Keys Botzum's Q&A
says the following:
The delegation chain of RPCs may cross only one cell boundary. This means
that with two cells you can not cross back into the originating cell,
and any delegation involving 3 cells will not work at all.

Q 2cc-07:: How can I find out who is currently logged in to a DCE cell?

You can't.

DCE does not track this information,
does not supply any tools to help collect this information,
does not provide any hooks in the security server to allow you
to collect this information.

Answering the question of "who is logged in" is difficult,
in part because in the DCE architecture it is hard to decide exactly
what the question means.
For example, if a person is logged in to a DCE client machine,
but his DCE tickets have expired, does that person count as logged in
or not?

If you must track who is logged in, you'll need to build
your own service to maintain the information.
An obvious approach is to capture login and logout events and
report them to some central location; this requires that you be able
to instrument all login/logout commands on hosts within your cell.
Another approach would be a daemon that runs on each host and
periodically notices ticket files; there are security concerns with this
Or you could just run "rwho" :-)


Q 2s-01: Where can I find an ACL manager to include in my application?


DCE version 1.1 includes a basic ACL manager library.

HP's OODCE (DCE C++ class library) provides you with a default,
built-in ACL Manager with the server class.

Transarc's Encina (transaction monitor) product includes
an ACL manager.

Lockhart's book contains a sample ACL manager. (see Q 3.02)

Hu's book on DCE Security Programming contains a sample ACL manager.
(see Q.3.02)

The code from both the Lockhart and Hu books is available on-line.
See Q 3.07 for URLs.

includes sample code for a greet-derived client/server pair where the
server uses the DCE 1.1 ACL manager library

Q 2s-02: Does DCE Security interoperate with other Kerberos systems?


Basically, no, or maybe yes, depending on what you want to do.

To use authenticated DCE services, you must have credentials from
the DCE security service; vanilla Kerberos v5 tickets aren't sufficient.
But then, to use DCE services you must be using DCE RPC, so this
is not really a problem.

Going the other way, it is expected that a DCE security server
can issue tickets that can be used by vanilla Kerberos applications.
The Open Group was wary of promising this until the Kerberos v5 specs were
published, but now that the Kerberos RFC has been published, Open Group
anticipates guaranteeing interoperability sometime "soon".

In a little more detail, the way to think about this is as follows:

Kerberos offers 2 services (Authentication Service, Ticket
Granting Service) over 1 communication mechanism (UDP port 88).

DCE security offers 3 services (AS, TGS, Privilege Service) over
2 communication mechanisms (UDP port 88, RPC).

Where Kerberos and DCE security intersect (AS, TGS over UDP port
88), the services are identical.

DCE V1.1 supports the GSSAPI, so non-DCE services that use GSSAPI
can be integrated with DCE security server.

Q 2s-03: Can I control the endpoints assigned to DCE servers?
For example, to allow DCE clients and servers to communicate
across a firewall?



You can, of course, assign well-known endpoints to all your servers,
but that is a truly bad idea, and won't work for services that you
don't have source code for (such as CDS and the security registry).

The endpoint mapper will examine the environment variable
RPC_RESTRICTED_PORTS and choose endpoints only in that range.
You can make the range as small as necessary, and configure your
firewall to pass traffic only to those ports.

The format of RPC_RESTRICTED_PORTS is a list of settings, separated by
colons. Each setting is of the form protocol[lo-hi]. E.g.

export RPC_RESTRICTED_PORTS=ncacn_ip_tcp[5000-6000]:ncadg_ip_udp[5000-6000]

Q 2s-04: Do DCE servers automatically update their long term secret keys?


Except for exceptional circumstances, all DCE servers should
periodically change their long-term key. However, neither the
servers provided by DCE nor those written by you or third
parties will do this out-of-the-box.

The way to have a server update its key is by spawning a
thread that calls sec_key_mgmt_manage_key() (which never returns under
normal circumstances). As distributed by Open Group, DCE has no password
expirations set, so sec_key_mgmt_manage_key() won't actually do anything.
You may set the password expiration time or lifespan using an admin
tool such as rgy_edit or dcecp.

In 1.0.x releases, DCE only enforced passwd expiration in the clients
(such as printing a warning in dce_login). As of DCE 1.1, however,
the security server will no longer grant a TGT for an account who's
password/key has expired, so servers that aren't correctly running the
manage key code before their password expires will require
administrative intervention to become operational again.

The following table shows what the Open Group-provided servers do:

cdsd -- yes; uses FILE:/krb5/v5srvtab
dced -- yes (uses FILE:/krb5/v5srvtab and manages hosts/<machine>/self)
dtsd -- not applicable; runs as the machine principal
pwd_strengthd -- yes; uses FILE:/krb5/pwd_strength_tab
secd (maintains three keys) -- not applicable

Q 2s-05:
What problems might I cause by changing the expiration time
or password lifespan in a running cell?

DCE v1.1 will not let you authenticate (acquire a TGT) with an expired
password. As such, if you change the "password lifespan" policy from
"never" or a period longer than 30 days, to days, then every host
principal, user and server account that has not (by coincidence) updated
their password/key within the last 30 days, will be locked out. Users
would not be able to log in, server applications would no longer be able
to obtain credentials and the dceds would no longer be able to
obtain/refresh their machine credentials.


Q2s-06: How do I begin to force password changes,
without getting into the trouble described in the previous question?

A: Set an explicit "expiration date" policy, some time in the near future.
(Make sure it's at least 30 minutes from the current time, to give
the key management threads time to catch the change and a couple of
chances to make the change.) At that specified date/time, all users,
server apps and machines should have updated their passwords/keys.
Recover from any problems that occurred (machines down during that
time, users forgetting...), then set the "password lifespan" policy
field to the desired limit.

Q 2s_07: Do all versions of DCE support the sec_key_mgmt_manage_key()
functionality in the same way?

A: Unfortunately not. DCE 1.0.x releases had a limitation in
this key mgmt call, in that it would not wake up until shortly before
the key was due to expire (based on the expiration data that it obtained
when the call was started or the last time through the loop after a
change). DCE 1.1 should no longer have this problem, as the call
wakes up every 10 minutes to see whether the expiration time has

Q 2s_08: How can I work around the key_mgmt problem if I have older
client machines, but DCE 1.1+ based servers?

A: Preferably, upgrade them all to an DCE 1.1 or later. If this is not possible,
manually change the keys for each older client:
rgy_edit-> kta -p hosts/<hostname>/self -a -r
(on each DCE 1.0.x-based client machine). Do this for any accounts in the
default keytable, and any servers running on older versions that also use
the sec_key_mgmt_manage_key() call.

Q 2c-01: Will Windows NT communicate with DCE?


Windows NT comes with an RPC which interoperates with DCE RPC.
Windows 95 apparently provides this interface as well.
However, it is not quite the same as DCE.

The wire-level protocol is the same as DCE RPC, so applications
running on NT can communicate with DCE applications on other
platforms. However, the application source code is not instantly
portable. Microsoft changed the format of procedure names
and moved the status result from a parameter to the function value.
This kind of change can be covered up by a set of preprocessor
macros, but it is a change to be dealt with.

A more serious consideration is that Microsoft's RPC does not
use the standard DCE services, such as directory service and security.
Thus applications that cross between Microsoft RPC and DCE will have
to make unauthenticated calls and use string bindings.
Digital bundles a Name Service Interface Daemon (nsid) into its
DCE products. The nsid fills in the gap and allows an NT client to
get bindings from CDS.

See the book "Distributing Applications Across DCE and Windows NT"
for more details (see Q 3.02).

The good news is that DEC and Gradient both have DCE
products for NT. Digital has a DCE client product available
for both Intel and Alpha architectures. Gradient has a full
secure core product.

Information on Digital's DCE on NT product is available at
Information from Gradient can be obtained by sending email to
or at their Web site:


Q 2c-02: Can I use DCE from C++?


Yes. First of all, since you can call C functions from C++
you can access all the DCE services from a C++ program. But
that will not give you the benefits of C++.

There are several packages that provide a C++ interface to DCE.
They different quite a bit in style and approach, so you'll
need to consider them carefully in light of your own needs and

Objtran was produced by Citibank, and is available by anonymous
ftp at ftp://wilma.cs.brown.edu/pub/Objtran.tar.Z

Hal Computer Systems developed a set of classes called DCE++.
That part of the company was spun off into Chisholm Technologies.
For information on DCE++, see their web page:

Hewlett-Packard has a product called OODCE. It is for sale,
and currently supported only on HP-UX. For technical information
retrieve the OSF RFC #49 (see Q 3.05); for sales
information contact your HP sales office.

DCE version 1.2.1 will include IDL support for C++. The OSF (oops,
the Open Group) recently announced that OODCE will be adopted as
part of DCE in a future version.

Transarc's Encina v2.0 includes Encina++, C++ support for Encina and DCE.


Q 2c-03: Can I write an application that uses DCE and X11/Motif?


Yes but there are several serious pitfalls.

The X11/Xt/Motif libraries may not be thread-safe. For example,
suppose one thread calls a function in Xt, which calls a nonthreadsafe
malloc(), which then gets preempted. The next thread may call
a threadsafe malloc() that comes with DCE. When control returns to
the first malloc(), any assumptions about the state of the heap are

Also, Motif/Xt/Xlib are not currently reentrant wrt/themselves. You can't
have multiple threads concurrently manipulating any Motif/Xt/Xlib
global state. Fortunately this issue is under you control when
designing the application. X11R6 includes a thread-safe version of
Xlib, but it will be a while yet before the vendors are all delivering
thread-safe Motif.

A related issue is that XtAppMainLoop() waits in a select() for activity,
coupled with the fact that DCE also waits in a select() for activity.
Unless the two are select()s are cooperating, one or the other will be
starved. This is a platform-specific issue, you should check with your
DCE vendor for full details.
If it is a problem in your environment, the standard solution is to
encapsulate the GUI in one process, the DCE client code in another process,
and connect them with a simple IPC such as a Unix pipe.

Q 2c-04: Is DCE RPC compatible with ONC RPC?


No. DCE and ONC both use the concept of the Remote Procedure Call,
but the wire protocols that they use are not compatible. You will
need to use either DCE for both client and server, or ONC for
both client and server; both products are available for most platforms.

It is possible for a single program to use both DCE and ONC.
Thus a server could be built to server both DCE and ONC clients,
or a gateway could be built to accept one kind of RPC and forward
to a server of another kind.

Q 2c-05: Is XDR the same as, or compatible with, NDR?
If not, how do I convert from one to the other?


(For those of you not fluent in three-letter abbreviations,
XDR is eXternal Data Representation, the data format used on the wire
by ONC, and NDR is Network Data Representation, used on the wire by DCE.)
XDR and NDR use different methods for tagging data items.


Tier 3 -- Data Management

Q 3.01: Where can I get online information about DCE


First of all, the official documentation for DCE is not available
for anonymous ftp. But there are several sites providing
lots of other information. Here are some of the most useful:

The Open Software Foundation maintains a WWW server with information
about all the Open Group products, including DCE.
takes you directly to the DCE index.
Highlights of the Open Group's web server include a complete list of DCE
products, the DCE Request for Comments documents (RFCs), the
DCE product catalog (see Q 1.04) and a
hypertext version of the Frequently Asked Questions list.

Project Pilgrim at the University of Massachussetts has a DCE homepage at
Their server provides a complete set of RFCS, searchable by http
and also by gopher, at
and a directory of contributed software, currently consisting of
performance measurement utilities from Pilgrim, and book examples from

The Center for Information Technology Integration at the
University of Michigan has several pieces of DCE information
on their servers. The big attraction here is the set of
CITI Technical Reports related to DCE.
You can get to the tech-reports from the CITI web server,

Transarc has some hints and examples available on the web.
These are informal, practical discussions of how to handle real-world
problems such as changing the IP address of the CDS server,
implementing peer-to-peer RPC, handling key management, etc.

Keys Botzum maintains an information page at
Look for the link to his personal question and answer page, which contains
DCE and Encina information.

Intellisoft's advutorial "DCE and The Enterprise"
is on-line and available via Intellisoft's web site

The PC Webopaedia http://www.sandybay.com/pc-web
contains pointers to number of DCE-related web pages (including this FAQ),
including magazine articles. Use the search feature to find DCE links.

IBM has a support web page for DCE on AIX at

Q 3.02: What books are published on DCE?


Documentation on DCE should be suppiled with vendor products. The
Open Group sells complete sets of documentation. The DCE set consists of
14 volumes and costs $525. Documentation for version 1.1 is
The three volumes of specifications (AES) can be purchased separately
for $100 (plus shipping).

Order documentation in the US by contacting:

OSF??? Direct Channels
Phone: 617-621-7300
E-mail: dir...@osf.org

Contact OSF Direct in Europe, East Asia, and Africa
at OSF's Brussels Office

Open Software Foundation
Avenue des Pleiades-laan 15
1200 Brussels
Phone: +32 2 772 8888, fax: +32 2 772 9228

Contact OSF Direct in the Pacific region at

Open Software Foundation
11-10, Kita-Aoyama 2-chrome
Minato-ku, Tokyo 107
Phone +813-3479-4740; fax: +813-3479-4760

The DCE documentation is also published by Prentice-Hall. These
books contain about the same material as the Open Group manuals, but
are edited to improve the presentation.

Introduction to DCE ISBN 0-13-490624-1
DCE Administration Reference ISBN 0-13-643818-0
DCE User's Guide and Reference ISBN 0-13-643842-3
DCE Application Development Guide ISBN 0-13-643826-1
DCE Application Development Reference ISBN 0-13-643834-2
DCE Administration Guide
Vol. 1, Introduction ISBN 0-13-176546-9
Vol. 2, Core Components ISBN 0-13-176553-1
Vol. 3, Extended Services ISBN 0-13-176561-2
Application Environment Specification/Distributed Computing
RPC Volume ISBN 0-13-043688-7

Other books on DCE:

Practical DCE Programming
by Charles Knouse (Hewlett Packard)
Prentice Hall ISBN 0-13-324419-9
Understanding OSF DCE 1.1 for AIX and OS/2
by Rolf Lendenmann.
Prentice Hall, ISBN: 0-13-493750-3 (paper)
Understanding DCE, by Rosenberry, Kenney, and Fisher
O'Reilly, ISBN 1-56592-005-8
Guide to writing DCE Appplications,2nd edition
by Shirley, Hu, and Magid
O'Reilly, ISBN 1-56592-045-7
Distributing Applications Across DCE and Windows NT,
by Rosenberry and Teague
O'Reilly, ISBN 1-56592-047-3
DCE Security
by Wei Hu
O'Reilly, ISBN 1-56592-134-8
DCE: A Guide to Developing Portable Applications
Michael T. Peterson
McGraw/Hill (Ranade Workstation Series)
ISBN 0079118003 (hard)
ISBN 0079118011 (paper)
OSF DCE: Guide to Developing Distributed Applications,
by H. W. Lockhart, Jr,
McGraw-Hill,(Ranade Workstation Series)
ISBN 0-07-911481-4
DCE--The OSF Distributed Computing Environment, Client/Server Model
and Beyond; Proceedings of the International Workshop on DCE, 1993
Lecture Notes in Computer Science #731, Springer-Verlag,
ISBN 3-540-57306-2

And for you German-speaking DCE-ers:

DCE--Das OSF Distributed Computing Environment, Einfuerhrung und Grundlagen,
by Alexander Schill, Springer-Verlag, ISBN 3-540-55335-5

Q 3.03: Where can I get more information about DCE and DCE-related products?


With WWW, access the URL: "http://www.opengroup.org/tech/dce/

The general response to any query of the form "Where can I get a _____ for
DCE?" is "Contact
dir...@osf.org" (phone +1 617 621-7300).

This mail list will reach the people at Open Group that maintain contacts with
just about everbody that has a DCE product to sell. You can get a listing
of the vendors and their products. These include DCE RPC debugging
environments, C++ toolkits, Visual-BASIC development environments, etc.

This mail list also reaches the people at Open Group that maintain the
lists of documentation available, both from OSF, and from outside OSF,
about DCE (and about Motif, DME, and OSF/1).

Transarc maintains a mail alias,
info...@transarc.com, which carries
discussions about DCE. Send to info-dce...@transarc.com to join the list.
(Transarc also has a list for Encina:
For information on Transarc products, send
mail to t...@transarc.com.

For Users of DCE on MVS, there is a mailing list:


and a Web page (under construction):

Q 3.04: What are DCE RFCs, and how can I get them?


DCE RFCs are requests for comments for ongoing DCE development. They
are similar in concept to the Internet RFCs. Nothing in there is
promised from by Open Group. They are a formal way to pass ideas among DCE
development partners.

You can access them by WWW (or gopher) by:



Q 3.05: Where can I get the Public Domain version of DCE?


In October 1994, Digital Equipment Corporation and Hewlett-Packard
released into the public domain the RPC implementation used by DCE.
This code includes the IDL compiler and the RPC runtime. It does
not include any of the other services: DTS, CDS, Security, DFS.
In fact, it is not a sufficient base for a client machine, as it
does not include the CDS, DTS, and security clerk processes that
are normally required.

The code is available on the Internet from several servers:




Be sure also to read the licensing information found in the
same directories.

Be warned that building anything from this release is not a simple
matter. DCE uses an older draft of the Posix threads standard,
the IDL compiler does some odd things with YACC, and there is no
simple configuration mechanism.

Andrew Sandoval has made a Linux port available. See:

Jim Doyle is working on porting all of DCE to Linux/FreeBSD. See:
Michael Peterson's work on DCE and Pthreads for Linux is available at:


Q 3.06: Is there a DCE Users Group I can join?


Yes. Besides the Open Group itself,
there are local DCE users groups in several areas.
Changes in the set of users groups have proven difficult
to track in this FAQ, so contact the Open Group to find the
user group nearest you.


Q 3.07: Are there any example programs available on-line?


The Open Group's web site includes the
Open Software Mall, and the Mall includes a list of

DCE-related free software. This is a good starting point for
a search. All of the following links are to be found in
the Mall.

The examples from Hal Lockhart's book are available at the OG Software
Mall. Note that you must start at the Mall homepage
and register before you can retrieve software for the first time.

The examples from the O'Reilly books (Shirley-Hu-Magid, Rosenberry-Teague,
Hu) are available at

Gradient Technologies has

sample PC-DCE code available at their ftp site.

Transarc's Web site has several code samples.

Tier 4 -- Acknowledgements

ACK: Thanks to the following, who provided netnews and email messages
from which this information is gleaned. All errors should be
blamed on Jon Mauney. (Corrections are actively solicited.)

Hal Lichtin, Open Software Foundation
Rich Salz, Open Software Foundation
Courtney Mark Grey, Open Software Foundation
Walt Tuvell Open Software Foundation
David Weisman Open Software Foundation
Matt Thomas, Digital Equipment Corporation
Jonathon Chinitz Intellisoft
Nat Mishkin Atria Software
Mark Hickey OpenVision
Keys Botzum Transarc
numerous others who have been accidentally omitted but
whose contributions are highly valued.

And special thanks to:

Harold Lockhart, Jr. Locus Computing Corporation

who wrote several of the answers in this FAQ.

<img src="http://www.sandybay.com/pc-web/images/awards/selected.gif"
width="68" height="63" border="0" alt="Selected by PC Webopaedia">

Reply all
Reply to author
0 new messages