Network Working Group S. Kille
Request for Comments: 1801 ISODE Consortium
Category: EXPerimental June 1995
X.400-MHS use of the X.500 Directory to support X.400-MHS Routing
Status of this Memo
This memo defines an Experimental Protocol for the Internet
community. This memo does not specify an Internet standard of any
kind. Discussion and suggestions for improvement are requested.
Distribution of this memo is unlimited.
Table of Contents
1 IntrodUCtion 3
2 Goals 3
3 Approach 5
4 Direct vs Indirect Connection 6
5 X.400 and RFC822 8
6 Objects 9
7 Communities 10
8 Routing Trees 11
8.1 Routing Tree Definition . . . . . . . 12
8.2 The Open Community Routing Tree . . . . . 12
8.3 Routing Tree Location . . . . . . . 13
8.4 Example Routing Trees . . . . . . . 13
8.5 Use of Routing Trees to look up Information . . 13
9 Routing Tree Selection 14
9.1 Routing Tree Order . . . . . . . . 14
9.2 Example use of Routing Trees . . . . . . 15
9.2.1 Fully Open Organisation . . . . . 15
9.2.2 Open Organisation with Fallback . . . 15
9.2.3 Minimal-routing MTA . . . . . . 16
9.2.4 Organisation with Firewall . . . . . 16
9.2.5 Well Known Entry Points . . . . . 16
9.2.6 ADMD using the Open Community for Advertising 16
9.2.7 ADMD/PRMD gateway . . . . . . . 17
10 Routing Information 17
10.1 Multiple routing trees . . . . . . . 20
10.2 MTA Choice . . . . . . . . . . 22
10.3 Routing Filters . . . . . . . . . 25
10.4 Indirect Connectivity . . . . . . . 26
11 Local Addresses (UAs) 27
11.1 Searching for Local Users . . . . . . 30
12 Direct Lookup 30
13 Alternate Routes 30
13.1 Finding Alternate Routes . . . . . . . 30
13.2 Sharing routing information . . . . . . 31
14 Looking up Information in the Directory 31
15 Naming MTAs 33
15.1 Naming 1984 MTAs . . . . . . . . . 35
16 Attributes Associated with the MTA 35
17 Bilateral Agreements 36
18 MTA Selection 38
18.1 Dealing with protocol mismatches . . . . . 38
18.2 Supported Protocols . . . . . . . . 39
18.3 MTA Capability Restrictions . . . . . . 39
18.4 SuBTree Capability Restrictions . . . . . 40
19 MTA Pulling Messages 41
20 Security and Policy 42
20.1 Finding the Name of the Calling MTA . . . . 42
20.2 Authentication . . . . . . . . . 42
20.3 Authentication Information . . . . . . 44
21 Policy and Authorisation 46
21.1 Simple MTA Policy . . . . . . . . 46
21.2 Complex MTA Policy . . . . . . . . 47
22 Delivery 49
22.1 Redirects . . . . . . . . . . 49
22.2 Underspecified O/R Addresses . . . . . . 50
22.3 Non Delivery . . . . . . . . . . 51
22.4 Bad Addresses . . . . . . . . . 51
23 Submission 53
23.1 Normal Derivation . . . . . . . . 53
23.2 Roles and Groups . . . . . . . . . 53
24 Access Units 54
25 The Overall Routing Algorithm 54
26 Performance 55
27 Acknowledgements 55
28 References 56
29 Security Considerations 57
30 Author"s Address 58
A Object Identifier Assignment 59
B Community Identifier Assignments 60
C Protocol Identifier Assignments 60
D ASN.1 Summary 61
E Regular Expression Syntax 71
List of Figures
1 Location of Routing Trees . . . . . . 12
2 Routing Tree Use Definition . . . . . . 14
3 Routing Information at a Node . . . . . 17
4 Indirect Access . . . . . . . . . 25
5 UA Attributes . . . . . . . . . 27
6 MTA Definitions . . . . . . . . . 33
7 MTA Bilateral Table Entry . . . . . . 36
8 Bilateral Table Attribute . . . . . . 37
9 Supported MTS Extensions . . . . . . . 39
10 Subtree Capability Restriction . . . . . 40
11 Pulling Messages . . . . . . . . . 41
12 Authentication Requirements . . . . . . 43
13 MTA Authentication Parameters . . . . . 45
14 Simple MTA Policy Specification . . . . . 46
15 Redirect Definition . . . . . . . . 48
16 Non Delivery Information . . . . . . . 50
17 Bad Address Pointers . . . . . . . . 52
18 Access Unit Attributes . . . . . . . 53
19 Object Identifier Assignment . . . . . . 59
20 Transport Community Object Identifier Assignments 60
21 Protocol Object Identifier Assignments . . . 61
22 ASN.1 Summary . . . . . . . . . 61
1. Introduction
MHS Routing is the problem of controlling the path of a message as it
traverses one or more MTAs to reach its destination recipients.
Routing starts with a recipient O/R Address, and parameters
associated with the message to be routed. It is assumed that this is
known a priori, or is derived at submission time as described in
Section 23.
The key problem in routing is to map from an O/R Address onto an MTA
(next hop). This shall be an MTA which in some sense is "nearer" to
the destination UA. This is done repeatedly until the message can be
directly delivered to the recipient UA. There are a number of things
which need to be considered to determine this. These are discussed
in the subsequent sections. A description of the overall routing
process is given in Section 25.
2. Goals
Application level routing for MHS is a complex procedure, with many
requirements. The following goals for the solution are set:
o Straightforward to manage. Non-trivial configuration of routing
for current message handling systems is a black art, often
involving gathering and processing many tables, and editing
complex configuration files. Many problems are solved in a very
ad hoc manner. Managing routing for MHS is the most serious
headache for most mail system managers.
o Economic, both in terms of network and computational resources.
o Robust. Errors and out of date information shall cause minimal
and localised damage.
o Deal with link failures. There needs to be some ability to choose
alternative routes. In general, it is desirable that the routing
approach be redundant.
o Load sharing. Information on routes shall allow "equal" routes
to be specified, and thus facilitate load sharing.
o Support format and protocol conversion
o Dynamic and automatic. There shall be no need for manual
propagation of tables or administrator intervention.
o Policy robust. It shall not allow specification of policies which
cause undesirable routing effects.
o Reasonably straightforward to implement.
o Deal with X.400, RFC822, and their interaction.
o Extensible to other mail architectures
o Recognise existing RFC822 routing, and coexist smoothly.
o Improve RFC822 routing capabilities. This is particularly
important for RFC822 sites not in the SMTP Internet.
o Deal correctly with different X.400 protocols (P1, P3, P7), and
with 1984, 1988 and 1992 versions.
o Support X.400 operation over multiple protocol stacks (TCP/IP,
CONS, CLNS) and in different communities.
o Messages shall be routed consistently. Alternate routing
strategies, which might introduce unexpected delay, shall be used
with care (e.g., routing through a protocol converter due to
unavailability of an MTA).
o Delay between message submission and delivery shall be minimised.
This has indirect impact on the routing approaches used.
o Interact sensibly with ADMD services.
o Be global in scope
o Routing strategy shall deal with a scale of order of magnitude
1,000,000 -- 100,000,000 MTAs.
o Routing strategy shall deal with of order 1,000,000 -- 100,000,000
Organisations.
o Information about alterations in topology shall propagate rapidly
to sites affected by the change.
o Removal, examination, or destruction of messages by third parties
shall be difficult. This is hard to quantify, but "difficult"
shall be comparable to the effort needed to break system security
on a typical MTA system.
o As with current Research Networks, it is recognised that
prevention of forged mail will not always be possible. However,
this shall be as hard as can be afforded.
o Sufficient tracing and logging shall be available to track down
security violations and faults.
o Optimisation of routing messages with multiple recipients, in
cases where this involves selection of preferred single recipient
routes.
The following are not initial goals:
o Advanced optimisation of routing messages with multiple
recipients, noting dependencies between the recipients to find
routes which would not have been chosen for any of the single
recipients.
o Dynamic load balancing. The approach does not give a means to
determine load. However, information on alternate routes is
provided, which is the static information needed for load
balancing.
3. Approach
A broad problem statement, and a survey of earlier approaches to the
problem is given in the COSINE Study on MHS Topology and Routing [8].
The interim (table-based) approach suggested in this study, whilst
not being followed in detail, broadly reflects what the research
X.400 (GO-MHS) community is doing. The evolving specification of the
RARE table format is defined in [5]. This document specifies the
envisaged longer term approach.
Some documents have made useful contributions to this work:
o A paper by the editor on MHS use of directory, which laid out the
broad approach of mapping the O/R Address space on to the DIT [7].
o Initial ISO Standardisation work on MHS use of Directory for
routing [19]. Subsequent ISO work in this area has drawn from
earlier drafts of this specification.
o The work of the VERDI Project [3].
o Work by Kevin Jordan of CDC [6].
o The routing approach of ACSNet [4, 17] paper. This gives useful
ideas on incremental routing, and replicating routing data.
o A lot of work on network routing is becoming increasingly
relevant. As the MHS routing problem increases in size, and
network routing increases in sophistication (e.g., policy based
routing), the two areas have increasing amounts in common. For
example, see [2].
4. Direct vs Indirect Connection
Two extreme approaches to routing connectivity are:
1. High connectivity between MTAs. An example of this is the way
the Domain Name Server system is used on the DARPA/NSF Internet.
Essentially, all MTAs are fully interconnected.
2. Low connectivity between MTAs. An example of this is the UUCP
network.
In general an intermediate approach is desirable. Too sparse a
connectivity is inefficient, and leads to undue delays. However,
full connectivity is not desirable, for the reasons discussed below.
A number of general issues related to relaying are now considered.
The reasons for avoiding relaying are clear. These include.
o Efficiency. If there is an open network, it is desirable that it
be used.
o Extra hops introduce delay, and increase the (very small)
possibility of message loss. As a basic principle, hop count
shall be minimised.
o Busy relays or Well Known Entry points can introduce high delay
and lead to single point of failure.
o If there is only one hop, it is straightforward for the user to
monitor progress of messages submitted. If a message is delayed,
the user can take appropriate action.
o Many users like the security of direct transmission. It is an
argument often given very strongly for use of SMTP.
Despite these very powerful arguments, there are a number of reasons
why some level of relaying is desirable:
o Charge optimisation. If there is an expensive network/link to be
traversed, it may make sense to restrict its usage to a small
number of MTAs. This would allow for optimisation with respect to
the charging policy of this link.
o Copy optimisation. If a message is being sent to two remote MTAs
which are close together, it is usually optimal to send the
message to one of the MTAs (for both recipients), and let it pass
a copy to the other MTA.
o To access an intermediate MTA for some value added service. In
particular for:
-- Message Format Conversion
-- Distribution List expansion
o Dealing with different protocols. The store and forward approach
allows for straightforward conversion. Relevant cases include:
-- Provision of X.400 over different OSI Stacks (e.g.,
Connectionless Network Service).
-- Use of a different version of X.400.
-- Interaction with non-X.400 mail services
o To compensate for inadequate directory services: If tables are
maintained in an ad hoc manner, the manual effort to gain full
connectivity is too high.
o To hide complexity of structure. If an organisation has many
MTAs, it may still be advantageous to advertise a single entry
point to the outside world. It will be more efficient to have an
extra hop, than to (widely) distribute the information required to
connect directly. This will also encourage stability, as
organisations need to change internal structure much more
frequently than their external entry points. For many
organisations, establishing such firewalls is high priority.
o To handle authorisation, charging and security issues. In
general, it is desirable to deal with user oriented authorisation
at the application level. This is essential when MHS specific
parameters shall be taken into consideration. It may well be
beneficial for organisations to have a single MTA providing access
to the external world, which can apply a uniform access policy
(e.g., as to which people are allowed access). This would be
particularly true in a multi-vendor environment, where different
systems would otherwise have to enforce the same policy --- using
different vendor-specific mechanisms.
In summary there are strong reasons for an intermediate approach.
This will be achieved by providing mechanisms for both direct and
indirect connectivity. The manager of a configuration will then be
able to make appropriate choices for the environment.
Two models of managing large scale routing have evolved:
1. Use of a global directory/database. This is the approach
proposed here.
2. Use of a routing table in each MTA, which is managed either by a
management protocol or by directory. This is coupled with means
to exchange routing information between MTAs. This approach is
more analogous to how network level routing is commonly performed.
It has good characteristics in terms of managing links and
dealing with link related policy. However, it assumes limited
connectivity and does not adapt well to a network environment
with high connectivity available.
5. X.400 and RFC822
This document defines mechanisms for X.400 message routing. It is
important that this can be integrated with RFC822 based routing, as
many MTAs will work in both communities. This routing document is
written with this problem in mind, and some work to verify this has
been done. support for RFC822 routing using the same basic
infrastructure is defined in a companion document [13]. In addition
support for X.400/RFC822 gatewaying is needed, to support
interaction. Directory based mechanisms for this are defined in
[16]. The advantages of the approach defined by this set of
specifications are:
o Uniform management for sites which wish to support both protocols.
o Simpler management for gateways.
o Improved routing services for RFC822 only sites.
For sites which are only X.400 or only RFC822, the mechanisms
associated with gatewaying or with the other form of addressing are
not needed.
6. Objects
It is useful to start with a manager"s perspective. Here is the set
of object classes used in this specification. It is important that
all information entered relates to something which is being managed.
If this is achieved, configuration decisions are much more likely to
be correct. In the examples, distinguished names are written using
the String Syntax for Distinguished Names [11]. The list of objects
used in this specification is:
User An entry representing a single human user. This will typically
be named in an organisational context. For example:
CN=Edgar Smythe,
O=Zydeco Services, C=GB
This entry would have associated information, such as telephone
number, postal address, and mailbox.
MTA A Message Transfer Agent. In general, the binding between
machines and MTAs will be complex. Often a small number of MTAs
will be used to support many machines, by use of local approaches
such as shared filestores. MTAs may support multiple protocols,
and will identify separate addressing information for each
protocol.
To achieve support for multiple protocols, an MTA is modelled as
an Application Process, which is named in the directory. Each MTA
will have one or more associated Application Entities. Each
Application Entity is named as a child of the Application Process,
using a common name which conveniently identifies the Application
Entity relative to the Application Process. Each Application
Entity supports a single protocol, although different Application
Entities may support the same protocol. Where an MTA only
supports one protocol or where the addressing information for all
of the protocols supported have different attributes to represent
addressing information (e.g., P1(88) and SMTP) the Application
Entity(ies) may be represented by the single Application Process
entry.
User Agent (Mailbox) This defines the User Agent (UA) to which mail
may be delivered. This will define the account with which the UA
is associated, and may also point to the user(s) associated with
the UA. It will identify which MTAs are able to access the UA.
(In the formal X.400 model, there will be a single MTA delivering
to a UA. In many practical configurations, multiple MTAs can
deliver to a single UA. This will increase robustness, and is
desirable.)
Role Some organisational function. For example:
CN=System Manager, OU=Sales,
O=Zydeco Services, C=GB
The associated entry would indicate the occupant of the role.
Distribution Lists There would be an entry representing the
distribution list, with information about the list, the manger,
and members of the list.
7. Communities
There are two basic types of agreement in which an MTA may participate
in order to facilitate routing:
Bilateral Agreements An agreement between a pair of MTAs to route
certain types of traffic. This MTA pair agreement usually
reflects some form of special agreement and in general bilateral
information shall be held for the link at both ends. In some
cases, this information shall be private.
Open Agreements An agreement between a collection of MTAs to behave
in a cooperative fashion to route traffic. This may be viewed as
a general bilateral agreement.
It is important to ensure that there are sufficient agreements in
place for all messages to be routed. This will usually be done by
having agreements which correspond to the addressing hierarchy. For
X.400, this is the model where a PRMD connects to an ADMD, and the
ADMD provides the inter PRMD connectivity, by the ability to route to
all other ADMDs. Other agreements may be added to this hierarchy, in
order to improve the efficiency of routing. In general, there may be
valid addresses, which cannot be routed to, either for connectivity
or policy reasons.
We model these two types of agreements as communities. A community
is a scope in which an MTA advertises its services and learns about
other services. Each MTA will:
1. Register its services in one or more communities.
2. Look up services in one or more communities.
In most cases an MTA will deal with a very small number of
communities --- very often one only. There are a number of different
types of community.
The open community This is a public/global scope. It reflects
routing information which is made available to any MTA which
wishes to use it.
The local community This is the scope of a single MTA. It reflects
routing information private to the MTA. It will contain an MTA"s
view of the set of bilateral agreements in which it participates,
and routing information private and local to the MTA.
Hierarchical communities A hierarchical community is a subtree of the
O/R Address tree. For example, it might be a management domain,
an organisation, or an organisational unit. This sort of
community will allow for firewalls to be established. A community
can have complex internal structure, and register a small subset
of that in the open community.
Closed communities A closed community is a set of MTAs which agrees
to route amongst themselves. Examples of this might be ADMDs
within a country, or a set of PRMDs representing the same
organisation in multiple countries.
Formally, a community indicates the scope over which a service is
advertised. In practice, it will tend to reflect the scope of
services offered. It does not make sense to offer a public service,
and only advertise it locally. Public advertising of a private
service makes more sense, and this is shown below. In general,
having a community offer services corresponding to the scope in which
they are advertised will lead to routing efficiency. Examples of how
communities can be used to implement a range of routing policies are
given in Section 9.2.
8. Routing Trees
Communities are a useful abstract definition of the routing approach
taken by this specification. Each community is represented in the
directory as a routing tree. There will be many routing trees
instantiated in the directory. Typically, an MTA will only be
registered in and make use of a small number of routing trees. In
most cases, it will register in and use the same set of routing
trees.
8.1 Routing Tree Definition
Each community has a model of the O/R address space. Within a
community, there is a general model of what to do with a given O/R
Address. This is structured hierarchically, according to the O/R
address hierarchy. A community can register different possible
actions, depending on the depth of match. This might include
identifying the MTA associated with a UA which is matched fully, and
providing a default route for an O/R address where there is no match
in the community --- and all intermediate forms. The name structure
of a routing tree follows the O/R address hierarchy, which is
specified in a separate document [15]. Where there is any routing
action associated with a node in a routing tree, the node is of
object class routingInformation, as defined in Section 10.
8.2 The Open Community Routing Tree
The routing tree of the open community starts at the root of the DIT.
This routing tree also serves the special function of instantiating
the global O/R Address space in the Directory. Thus, if a UA wishes
to publish information to the world, this hierarchy allows it to do
so.
The O/R Address hierarchy is a registered tree, which may be
instantiated in the directory. Names at all points in the tree are
valid, and there is no requirement that the namespace is instantiated
by the owner of the name. For example, a PRMD may make an entry in
the DIT, even if the ADMD above it does not. In this case, there
will be a "skeletal" entry for the ADMD, which is used to hang the
PRMD entry in place. The skeletal entry contains the minimum number
of entries which are needed for it to exist in the DIT (Object Class
and Attribute information needed for the relative distinguished
name). This entry may be placed there solely to support the
subordinate entry, as its existence is inferred by the subordinate
entry. Only the owner of the entry may place information into it.
An analogous situation in current operational practice is to make DIT
entries for Countries and US States.
---------------------------------------------------------------------
routingTreeRoot OBJECT-CLASS ::= {
SUBCLASS OF {routingInformationsubtree}
ID oc-routing-tree-root}
Figure 1: Location of Routing Trees
---------------------------------------------------------------------
8.3 Routing Tree Location
All routing trees follow the same O/R address hierarchy. Routing
trees other than the open community routing tree are rooted at
arbitrary parts of the DIT. These routing trees are instantiated
using the subtree mechanism defined in the companion document
"Representing Tables and Subtrees in the Directory" [15]. A routing
tree is identified by the point at which it is rooted. An MTA will
use a list of routing trees, as determined by the mechanism described
in Section 9. Routing trees may be located in either the
organisational or O/R address structured part of the DIT. All routing
trees, other than the open community routing tree, are rooted by an
entry of object class routingTreeRoot, as defined in Figure 1.
8.4 Example Routing Trees
Consider routing trees with entries for O/R Address:
P=ABC; A=XYZMail; C=GB;
In the open community routing tree, this would have a distinguished
name of:
PRMD=ABC, ADMD=XYZMail, C=GB
Consider a routing tree which is private to:
O=Zydeco Services, C=GB
They might choose to label a routing tree root "Zydeco Routing Tree",
which would lead to a routing tree root of:
CN=Zydeco Routing Tree, O=Zydeco Services, C=GB
The O/R address in question would be stored in this routing tree as:
PRMD=ABC, ADMD=XYZMail
C=GB, CN=Zydeco Routing Tree,
O=Zydeco Services, C=GB
8.5 Use of Routing Trees to look up Information
Lookup of an O/R address in a routing tree is done as follows:
1. Map the O/R address onto the O/R address hierarchy described in
[15] in order to generate a Distinguished Name.
2. Append this to the Distinguished Name of the routing tree, and
then look up the whole name.
3. Handling of errors will depend on the application of the lookup,
and is discussed later.
Note that it is valid to look up a null O/R Address, as the routing
tree root may contain default routing information for the routing
tree. This is held in the root entry of the routing tree, which is a
subclass of routingInformation. The open community routing tree does
not have a default.
Routing trees may have aliases into other routing trees. This will
typically be done to optimise lookups from the first routing tree
which a given MTA uses. Lookup needs to take account of this.
9. Routing Tree Selection
The list of routing trees which a given MTA uses will be represented
in the directory. This uses the attribute defined in Figure 2.
---------------------------------------------------------------------
routingTreeList ATTRIBUTE ::= {
WITH SYNTAX RoutingTreeList
SINGLE VALUE
ID at-routing-tree-list}
RoutingTreeList ::= SEQUENCE OF RoutingTreeName
RoutingTreeName ::= DistinguishedName
Figure 2: Routing Tree Use Definition
---------------------------------------------------------------------
This attribute defines the routing trees used by an MTA, and the
order in which they are used. Holding these in the directory eases
configuration management. It also enables an MTA to calculate the
routing choice of any other MTA which follows this specification,
provided that none of its routing trees have access restrictions.
This will facilitate debugging routing problems.
9.1 Routing Tree Order
The order in which routing trees are used will be critical to the
operation of this algorithm. A common approach will be:
1. Access one or more shared private routing trees to access private
routing information.
2. Utilise the open routing tree.
3. Fall back to a default route from one of the private routing
trees.
Initially, the open routing tree will be very sparse, and there will
be little routing information in ADMD level nodes. Access to many
services will only be via ADMD services, which in turn will only be
accessible via private links. For most MTAs, the fallback routing
will be important, in order to gain access to an MTA which has the
right private connections configured.
In general, for a site, UAs will be registered in one routing tree
only, in order to avoid duplication. They may be placed into other
routing trees by use of aliases, in order to gain performance. For
some sites, Users and UAs with a 1:1 mapping will be mapped onto
single entries by use of aliases.
9.2 Example use of Routing Trees
Some examples of how this structure might be used are now given.
Many other combinations are possible to suit organisational
requirements.
9.2.1 Fully Open Organisation
The simplest usage is to place all routing information in the open
community routing tree. An organisation will simply establish O/R
addresses for all of its UAs in the open community tree, each
registering its supporting MTA. This will give access to all systems
accessible from this open community.
9.2.2 Open Organisation with Fallback
In practice, some MTAs and MDs will not be directly reachable from
the open community (e.g., ADMDs with a strong model of bilateral
agreements). These services will only be available to
users/communities with appropriate agreements in place. Therefore it
will be useful to have a second (local) routing tree, containing only
the name of the fallback MTA at its root. In many cases, this
fallback would be to an ADMD connection.
Thus, open routing will be tried first, and if this fails the message
will be routed to a single selected MTA.
9.2.3 Minimal-routing MTA
The simplest approach to routing for an MTA is to deliver messages to
associated users, and send everything else to another MTA (possibly
with backup).
An organisation using MTAs with this approach will register its users
as for the fully open organisation. A single routing tree will be
established, with the name of the organisation being aliased into the
open community routing tree. Thus the MTA will correctly identify
local users, but use a fallback mechanism for all other addresses.
9.2.4 Organisation with Firewall
An organisation can establish an organisation community to build a
firewall, with the overall organisation being registered in the open
community. This is an important structure, which it is important to
support cleanly.
o Some MTAs are registered in the open community routing tree to
give access into the organisation. This will include the O/R tree
down to the organisational level. Full O/R Address verification
will not take place externally.
o All users are registered in a private (organisational) routing
tree.
o All MTAs in the organisation are registered in the organisation"s
private routing tree, and access information in the organisation"s
community. This gives full internal connectivity.
o Some MTAs in the organisation access the open community routing
tree. These MTAs take traffic from the organisation to the
outside world. These will often be the same MTAs that are
externally advertised.
9.2.5 Well Known Entry Points
Well known entry points will be used to provide access to countries
and MDs which are oriented to private links. A private routing tree
will be established, which indicates these links. This tree would be
shared by the well known entry points.
9.2.6 ADMD using the Open Community for Advertising
An ADMD uses the open community for advertising. It advertises its
existence and also restrictive policy. This will be useful for:
o Address validation
o Advertising the mechanism for a bilateral link to be established
9.2.7 ADMD/PRMD gateway
An MTA provides a gateway from a PRMD to an ADMD. It is important to
note that many X.400 MDs will not use the directory. This is quite
legitimate. This technique can be used to register access into such
communities from those that use the directory.
o The MTA registers the ADMD in its local community (private link)
o The MTA registers itself in the PRMD"s community to give access to
the ADMD.
10. Routing Information
Routing trees are defined in the previous section, and are used as a
framework to hold routing information. Each node, other than a
skeletal one, in a routing tree has information associated with it,
which is defined by the object class routingInformation in Figure 3.
This structure is fundamental to the operation of this specification,
and it is recommended that it be studied with care.
---------------------------------------------------------------------
routingInformation OBJECT-CLASS ::= {
SUBCLASS OF top
KIND auxiliary
MAY CONTAIN {
subtreeInformation
routingFilter
routingFailureAction
mTAInfo
accessMD 10
nonDeliveryInfo
badAddressSearchPoint
badAddressSearchAttributes}
ID oc-routing-information}
-- No naming attributes as this is not a
-- structural object class
subtreeInformation ATTRIBUTE ::= { 20
WITH SYNTAX SubtreeInfo
SINGLE VALUE
ID at-subtree-information}
SubtreeInfo ::= ENUMERATED {
all-children-present(0),
not-all-children-present(1) }
routingFilter ATTRIBUTE ::= { 30
WITH SYNTAX RoutingFilter
ID at-routing-filter}
RoutingFilter ::= SEQUENCE{
attribute-type OBJECT-IDENTIFIER,
weight RouteWeight,
dda-key String OPTIONAL,
regex-match IA5String OPTIONAL,
node DistinguishedName } 40
String ::= CHOICE {PrintableString, TeletexString}
routingFailureAction ATTRIBUTE ::= {
WITH SYNTAX RoutingFailureAction
SINGLE VALUE
ID at-routing-failure-action}
RoutingFailureAction ::= ENUMERATED {
next-level(0), 50
next-tree-only(1),
next-tree-first(2),
stop(3) }
mTAInfo ATTRIBUTE ::= {
WITH SYNTAX MTAInfo
ID at-mta-info}
MTAInfo ::= SEQUENCE { 60
name DistinguishedName,
weight [1] RouteWeight DEFAULT preferred-access,
mta-attributes [2] SET OF Attribute OPTIONAL,
ae-info SEQUENCE OF SEQUENCE {
aEQualifier PrintableString,
ae-weight RouteWeight DEFAULT preferred-access,
ae-attributes SET OF Attribute OPTIONAL} OPTIONAL
}
RouteWeight ::= INTEGER {endpoint(0), 70
preferred-access(5),
backup(10)} (0..20)
Figure 3: Routing Information at a Node
---------------------------------------------------------------------
For example, information might be associated with the (PRMD) node:
PRMD=ABC, ADMD=XYZMail, C=GB
If this node was in the open community routing tree, then the
information represents information published by the owner of the PRMD
relating to public access to that PRMD. If this node was present in
another routing tree, it would represent information published by the
owner of the routing tree about access information to the referenced
PRMD. The attributes associated with a routingInformation node
provide the following information:
Implicit That the node corresponds to a partial or entire valid O/R
address. This is implicit in the existence of the entry.
Object Class If the node is a UA. This will be true if the node is of
object class routedUA. This is described further in Section 11.
If it is not of this object class, it is an intermediate node in
the O/R Address hierarchy.
routingFilter A set of routing filters, defined by the routingFilter
attribute. This attribute provides for routing on information in
the unmatched part of the O/R Address. This is described in
Section 10.3.
subtreeInformation Whether or not the node is authoritative for the
level below is specified by the subtreeInformation attribute. If
it is authoritative, indicated by the value all-children-present,
this will give the basis for (permanently) rejecting invalid O/R
Addresses. The attribute is encoded as enumerated, as it may be
later possible to add partial authority (e.g., for certain
attribute types). If this attribute is missing, the node is
assumed to be non-authoritative (not-all-children-present).
The value all-children-present simply means that all of the child
entries are present, and that this can be used to determine
invalid addresses. There are no implications about the presence
of routing information. Thus it is possible to verify an entire
address, but only to route on one of the higher level components.
For example, consider the node:
MHS-O=Zydeco, PRMD=ABC, ADMD=XYZMail, C=GB
An organisation which has a bilateral agreement with this
organisation has this entry in its routing tree, with no children
entries. This is marked as non-authoritative. There is a second
routing tree maintained by Zydeco, which contains all of the
children of this node, and is marked as authoritative. When
considering an O/R Address
MHS-G=Random + MHS-S=Unknown, MHS-O=Zydeco,
PRMD=ABC, ADMD=XYZMail, C=GB
only the second, authoritative, routing tree can be used to
determine that this address is invalid. In practice, the manager
configuring the non-authoritative tree, will be able to select
whether an MTA using this tree will proceed to full verification,
or route based on the partially verified information.
mTAInfo A list of MTAs and associated information defined by the
mTAInfo attribute. This information is discussed further in
Sections 15 and 18. This information is the key information
associated with the node. When a node is matched in a lookup, it
indicates the validity of the route, and a set of MTAs to connect
to. Selection of MTAs is discussed in Sections 18 and
Section 10.2.
routingFailureAction An action to be taken if none of the MTAs can be
used directly (or if there are no MTAs present) is defined by the
routingFailureAction attribute. Use of this attribute and
multiple routing trees is described in Section 10.1.
accessMD The accessMD attribute is discussed in Section 10.4. This
attribute is used to indicate MDs which provide indirect access
to the part of the tree that is being routed to.
badAddressSearchPoint/badAddressSearchAttributes The
badAddressSearchPoint and badAddressSearchAttributes are
discussed in Section 17. This attribute is for when an address
has been rejected, and allows information on alternative addresses
to be found.
10.1 Multiple routing trees
A routing decision will usually be made on the basis of information
contained within multiple routing trees. This section describes the
algorithms relating to use of multiple routing trees. Issues
relating to the use of X.500 and handling of errors is discussed in
Section 14. The routing decision works by examining a series of
entries (nodes) in one or more routing trees. This information is
summarised in Figure 3. Each entry may contain information on
possible next-hop MTAs. When an entry is found which enables the
message to be routed, one of the routing options determined at this
point is selected, and a routing decision is made. It is possible
that further entries may be examined, in order to determine other
routing options. This sort of heuristic is not discussed here.
When a single routing tree is used, the longest possible match based
on the O/R address to be routed to is found. This entry, and then
each of its parents in turn is considered, ending with the routing
tree root node (except in the case of the open routing tree, which
does not have such a node). When multiple routing trees are
considered, the basic approach is to treat them in a defined order.
This is supplemented by a mechanism whereby if a matched node cannot
be used directly, the routing algorithm will have the choice to move
up a level in the current routing tree, or to move on to the next
routing tree with an option to move back to the first tree later.
This option to move back is to allow for the common case where a tree
is used to specify two things:
1. Routing information private to the MTA (e.g., local UAs or routing
info for bilateral links).
2. Default routing information for the case where other routing has
failed.
The actions allow for a tree to be followed, for the private
information, then for other trees to be used, and finally to fall
back to the default situation. For very complex configurations it
might be necessary to split this into two trees. The options defined
by routingFailureAction, to be used when the information in the entry
does not enable a direct route, are:
next-level Move up a level in the current routing tree. This is the
action implied if the attribute is omitted. This will usually be
the best action in the open community routing tree.
next-tree-only Move to the next tree, and do no further processing on
the current tree. This will be useful optimisation for a routing
tree where it is known that there is no useful additional routing
information higher in the routing tree.
next-tree-first Move to the next tree, and then default back to the
next level in this tree when all processing is completed on
subsequent trees. This will be useful for an MTA to operate in
the sequence:
1. Check for optimised private routes
2. Try other available information
3. Fall back to a local default route
stop This address is unroutable. No processing shall be done in any
trees.
For the root entry of a routing tree, the default action and next-
level are interpreted as next-tree-only.
10.2 MTA Choice
This section considers how the choice between alternate MTAs is made.
First, it is useful to consider the conditions why an MTA is entered
into a node of the routing tree:
o The manager for the node of the tree shall place it there. This
is a formality, but critical in terms of overall authority.
o The MTA manager shall agree to it being placed there. For a well
operated MTA, the access policy of the MTA will be set to enforce
this.
o The MTA will in general (for some class of message) be prepared
to route to any valid O/R address in the subtree implied by the
address. The only exception to this is where the MTA will route
to a subset of the tree which cannot easily be expressed by
making entries at the level below. An example might be an MTA
prepared to route to all of the subtree, with certain explicit
exceptions.
Information on each MTA is stored in an mTAInfo attribute, which is
defined in Figure 3. This attribute contains:
name The Distinguished Name of the MTA (Application Process)
weight A weighting factor (Route Weight) which gives a basis to
choose between different MTAs. This is described in Section 10.2.
mta-attributes Attributes from the MTA"s entry. Information on the
MTA will always be stored in the MTA"s entry. The MTA is
represented here as a structure, which enables some of this entry
information to be represented in the routing node. This is
effectively a maintained cache, and can lead to considerable
performance optimisation. For example if ten MTAs were
represented at a node, another MTA making a routing decision might
need to make ten directory reads in order to obtain the
information needed. If any attributes are present here, all of
the attributes needed to make a routing decision shall be
included, and also all attributes at the Application Entity level.
ae-info Where an MTA supports a single protocol only, or the
protocols it supports have address information that can be
represented in non-conflicting attributes, then the MTA may be
represented as an application process only. In this case, the
ae-info structure which gives information on associated
application entities may be omitted, as the MTA is represented by
a single application entity which has the same name as the
application process. In other cases, the names of all application
entities shall be included. A weight is associated with each
application entity to allow the MTA to indicate a preference
between its application entities.
The structure of information within ae-info is as follows:
ae-qualifier A printable string (e.g., "x400-88"), which is the
value of the common name of the relative distinguished name of the
application entity. This can be used with the application process
name to derive the application entity title.
ae-weight A weighting factor (Route Weight) which gives a basis to
choose between different Application Entities (not between
different MTAs). This is described below.
ae-attributes Attributes from the AEs entry.
Information in the mta-attributes and ae-info is present as a
performance optimisation, so that routing choices can be made with a
much smaller number of directory operations. Using this information,
whose presence is optional, is equivalent to looking up the
information in the MTA. If this information is present, it shall be
maintained to be the same as that information stored in the MTA
entry. Despite this maintenence requirement, use of this performance
optimisation data is optional, and the information may always be
looked up from the MTA entry.
Note: It has been suggested that substantial performance optimisation
will be achieved by caching, and that the performance gained
from maintaining these attributes does not justify the effort
of maintaining the entries. If this is borne out by
operational experience, this will be reflected in future
versions of this specification.
Route weighting is a mechanism to distinguish between different route
choices. A routing weight may be associated with the MTA in the
context of a routing tree entry. This is because routing weight will
always be context dependent. This will allow machines which have
other functions to be used as backup MTAs. The Route Weight is an
integer in range 0--20. The lower the value, the better the choice
of MTA. Where the weight is equal, and no other factors apply, the
choice between the MTAs shall be random to facilitate load balancing.
If the MTA itself is in the list, it shall only route to an MTA of
lower weight. The exact values will be chosen by the manager of the
relevant part of the routing tree. For guidance, three fixed points
are given:
o 0. For an MTA which can deliver directly to the entire subtree
implied by the position in the routing tree.
o 5. For an MTA which is preferred for this point in the subtree.
o 10. For a backup MTA.
When an organisation registers in multiple routing trees, the route
weight used is dependent on the context of the subtree. In general
it is not possible to compare weights between subtrees. In some
cases, use of route weighting can be used to divert traffic away from
expensive links.
Attributes present in an MTA Entry are defined in various parts of
this specification. A summary and pointers to these sections is
given in Section 16.
Attributes that are available in the MTA entry and will be needed for
making a routing choice are:
protocolInformation
applicationContext
mhs-deliverable-content-length
responderAuthenticationRequirements
initiatorAuthenticationRequirements
responderPullingAuthenticationRequirements
initiatorPullingAuthenticationRequirements
initiatorP1Mode
responderP1Mode
polledMTAs Current MTA shall be in list if message is to be pulled.
mTAsAllowedToPoll
supportedMTSExtensions
If any MTA attributes are present in the mTAInfo attribute, all of
the attributes that may affect routing choice shall be present.
Other attributes may be present. A full list of MTA attributes, with
summaries of their descriptions are given in Section 16, with a
formal definition in Figure 6.
10.3 Routing Filters
This attribute provides for routing on information in the unmatched
part of the O/R Address, including:
o Routing on the basis of an O/R Address component type
o Routing on the basis of a substring match of an O/R address
component. This might be used to route X121 addressed faxes to
an appropriate MTA.
When present, the procedures of analysing the routing filters shall
be followed before other actions. The routing filter overrides
mTAInfo and accessMD attributes, which means that the routing filter
must be considered first. Only in the event that no routing filters
match shall the mTAInfo and accessMD attributes be considered. The
components of the routingFilter attribute are:
---------------------------------------------------------------------
attribute-type This gives the attribute type to be matched, and is
selected from the attribute types which have not been matched to
identify the routing entry. The filter applies to this attribute
type. If there is no regular expression present (as defined
below), the filter is true if the attribute is present. The
value is the object identifier of the X.500 attribute type
(e.g., at-prmd-name).
weight This gives the weight of the filter, which is encoded as a
Route Weight, with lower values indicating higher priority. If
multiple filters match, the weight of each matched filter is used
to select between them. If the weight is the same, then a random
choice shall be made.
dda-key If the attribute is domain defined, then this parameter may
be used to identify the key.
accessMD ATTRIBUTE ::= {
SUBTYPE OF distinguishedName
ID at-access-md}
Figure 4: Indirect Access
---------------------------------------------------------------------
regex-match This string is used to give a regular expression match on
the attribute value. The syntax for regular expressions is
defined in Appendix E.
node This distinguished name specifies the entry which holds routing
information for the filter. It shall be an entry with object
class routingInformation, which can be used to determine the MTA
or MTA choice. All of the attributes from this entry should be
used, as if they had been directly returned from the current entry
(i.e., the procedure recurses). The current entry does not set
defaults.
An example of use of routing filters is now given, showing how to
route on X121 address to a fax gateway in Germany. Consider the
routing point.
PRMD=ABC, ADMD=XYZMail, C=GB
The entry associated would have two routing filters:
1. One with type x121 and no regular expression, to route a default
fax gateway.
2. One with type x121 and a regular expression ^9262 to route all
German faxes to a fax gateway located in Germany with which there
is a bilateral agreement. This would have a lower weight, so that
it would be selected over the default fax gateway.
10.4 Indirect Connectivity
In some cases a part of the O/R Address space will be accessed
indirectly. For example, an ADMD without access from the open
community might have an agreement with another MD to provide this
access. This is achieved by use of the accessMD attribute defined in
Figure 4. If this attribute is found, the routing algorithm shall
read the entry pointed to by this distinguished name. It shall be an
entry with object class routingInformation, which can be used to
determine the MTA or MTA choice and route according to the
information retrieve to this access MD. All of the attributes from
this entry should be used, as if they had been directly returned from
the current entry (i.e., the procedure recurses). The current entry
does not set defaults.
The attribute is called an MD, as this is descriptive of its normal
use. It might point to a more closely defined part of the O/R
Address space.
It is possible for both access MD and MTAs to be specified. This
might be done if the MTAs only support access over a restricted set
of transport stacks. In this case, the access MD shall only be
routed to if it is not possible to route to any of the MTAs.
This structure can also be used as an optimisation, where a set of
MTAs provides access to several parts of the O/R Address space.
Rather than repeat the MTA information (list of MTAs) in each
reference to the MD, a single access MD is used as a means of
grouping the MTAs. The value of the Distinguished Name of the access
MD will probably not be meaningful in this case (e.g., it might be
the name "Access MTA List", within the organisation.)
If the MTA routing is unable to access the information in the Access
MD due to directory security restrictions, the routing algorithm
shall continue as if no MTA information was located in the routing
entry.
11. Local Addresses (UAs)
Local addresses (UAs) are a special case for routing: the endpoint.
The definition of the routedUA object class is given in Figure 5.
This identifies a User Agent in a routing tree. This is needed for
several reasons:
---------------------------------------------------------------------
routedUA OBJECT-CLASS ::= {
SUBCLASS OF {routingInformation}
KIND auxiliary
MAY CONTAIN {
-- from X.402
mhs-deliverable-content-length
mhs-deliverable-content-types
mhs-deliverable-eits
mhs-message-store 10
mhs-preferred-delivery-methods
-- defined here
supportedExtensions
redirect
supportingMTA
userName
nonDeliveryInfo}
ID oc-routed-ua}
supportedExtensions ATTRIBUTE ::= { 20
SUBTYPE OF objectIdentifier
ID at-supported-extensions}
supportingMTA ATTRIBUTE ::= {
SUBTYPE OF mTAInfo
ID at-supporting-mta}
userName ATTRIBUTE ::= {
SUBTYPE OF distinguishedName
ID at-user-name} 30
Figure 5: UA Attributes
---------------------------------------------------------------------
1. To allow UAs to be defined without having an entry in another part
of the DIT.
2. To identify which (leaf and non-leaf) nodes in a routing tree are
User Agents. In a pure X.400 environment, a UA (as distinct from
a connecting part of the O/R address space) is simply identified
by object class. Thus an organisation entry can itself be a UA. A
UA need not be a leaf, and can thus have children in the tree.
3. To allow UA parameters as defined in X.402 (e.g., the
mhs-deliverable-eits) to be determined efficiently from the
routing tree, without having to go to the user"s entry.
4. To provide access to other information associated with the UA, as
defined below.
The following attributes are defined associated with the UA.
supportedExtensions MTS extensions supported by the MTA, which affect
delivery.
supportingMTA The MTAs which support a UA directly are noted in the
supportingMTA attribute, which may be multi-valued. In the X.400
model, only one MTA is associated with a UA. In practice, it is
possible and useful for several MTAs to be able to deliver to a
single UA. This attribute is a subtype of mTAInfo, and it defines
access information for an MTA which is able to deliver to the UA.
There may also be an mTAInfo attribute in the entry.
Components of the supportingMTA attribute are interpreted in the
same manner as mtaInfo is for routing, with one exception. The
values of the Route Weight are interpreted in the following
manner:
o 0. A preferred MTA for delivery.
o 5. A backup MTA.
o 10. A backup MTA, which is not presferred.
The supportingMTA attribute shall be present, unless the address
is being non-delivered or redirected, in which case it may be
omitted.
redirect The redirect attribute controls redirects, as described in
Section 22.1.
userName The attribute userName points to the distinguished Name of
the user, as defined by the mhs-user in X.402. The pointer from
the user to the O/R Address is achieved by the mhs-or-addresses
attribute. This makes the UA/User linkage symmetrical.
nonDeliveryInfo The attribute nonDeliveryInfo mandates non-delivery
to this address, as described in Section 22.3.
When routing to a UA, an MTA will read the supportingMTA attribute.
If it finds its own name present, it will know that the UA is local,
and invoke appropriate procedures for local delivery (e.g., co-
resident or P3 access information). The cost of holding these
attributes for each UA at a site will often be reduced by use of
shared attributes (as defined in X.500(93)).
Misconfiguration of the supportingMTA attribute could have serious
operational and possibly security problems, although for the most
part no worse than general routing configuration problems. An MTA
using this attribute may choose to perform certain sanity checks,
which might be to verify the routing tree or subtree that the entry
resides in.
The linkage between the UA and User entries was noted above. It is
also possible to use a single entry for both User and UA, as there is
no conflict between the attributes in each of the objects. In this
case, the entries shall be in one part of the DIT, with aliases from
the other. Because the UA and User are named with different
attributes, the aliases shall be at the leaf level.
11.1 Searching for Local Users
The approach defined in this specification performs all routing by
use of reads. This is done for performance reasons, as it is a
reasonable expectation that all DSA implementations will support a
high performance read operation. For local routing only, an MTA in
cooperation with the provider of the local routing tree may choose to
use a search operation to perform routing. The major benefit of this
is that there will not be a need to store aliases for alternate
names, and so the directory storage requirement and alias management
will be reduced. The difficulty with this approach is that it is
hard to define search criteria that would be effective in all
situations and well supported by all DUAs. There are also issues
about determining the validity of a route on the basis of partial
matches.
12. Direct Lookup
Where an O/R address is registered in the open community and has one
or more "open" MTAs which support it, this will be optimised by
storing MTA information in the O/R address entry. In general, the
Directory will support this by use of attribute inheritance or an
implementation will optimise the storage or repeated information, and
so there will not be a large storage overhead implied. This is a
function of the basic routing approach. As a further optimisation of
this case, the User"s distinguished name entry may contain the
mTAInfo attribute. This can be looked up from the distinguished
name, and thus routing on submission can be achieved by use of a
single read.
Note: This performance optimisation has a management overhead, and
further experience is needed to determine if the effort
justifies the performance improvement.
13. Alternate Routes
13.1 Finding Alternate Routes
The routing algorithm selects a single MTA to be routed to. It could
be extended to find alternate routes to a single MTA with possibly
different weights. How far this is done is a local configuration
choice. Provision of backup routing is desirable, and leads to
robust service, but excessive use of alternate routing is not usually
beneficial. It will often force messages onto convoluted paths, when
there was only a short outage on the preferred path. It is important
to note that this strategy will lead to picking the first acceptable
route. It is important to configure the routing trees so that the
first route identified will also be the best route.
13.2 Sharing routing information
So far, only single addresses have been considered. Improving
routing choice for multiple addresses is analogous to dealing with
multiple routes. This section defines an optional improvement. When
multiple addresses are present, and alternate routes are available,
the preferred routes may be chosen so as to maximise the number of
recipients sent with each message.
Specification of routing trees can facilitate this optimisation.
Suppose there is a set of addresses (e.g., in an organisation) which
have different MTAs, but have access to an MTA which will do local
switching. If each address is registered with the optimal MTA as
preferred, but has the "hub" MTA registered with a higher route
weight, then optimisation may occur when a message is sent to
multiple addresses in the group.
14. Looking up Information in the Directory
The description so far has been abstract about lookup of information.
This section considers how information is looked up in the Directory.
Consider that an O/R Address is presented for lookup, and there is a
sequence of routing trees. At any point in the lookup sequence,
there is one of a set of actions that can take place:
Entry Found Information from the entry (node) is returned and shall
be examined. The routing process continues or terminates, based
on this information.
Entry Not Found Return information on the length of best possible
match to the routing algorithm.
Temporary Reject The MTA shall stop the calculation, and repeat the
request later. Repeated temporary rejects should be handled in a
similar manner to the way the local MTA would handle the failure
to connect to a remote MTA.
Permanent Reject Administrative error on the directory which may be
fixed in future, but which currently prevents routing.