Using the Tuxedo Domains Component
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
The following sections explain how to administer a BEA Tuxedo Domains environment:
To integrate the Domains component with an existing BEA Tuxedo application, you add entries for domain gateway groups and gateway servers to the TUXCONFIG
file. You can use either the tmconfig(1)
or tmadmin(1)
command to add a Domains configuration to a running BEA Tuxedo application. You can also use tmadmin
to list the information available in the bulletin board for domain gateway groups and individual domain gateways.
Once your Domains environment is configured and integrated, you can administer it dynamically using a set of administrative tools provided by the Domains component. For example, you can specify and modify the list of services that are accessible across applications. The Domains software preserves the characteristics of the BEA Tuxedo programming interface (ATMI) and extends the scope of the ATMI so that clients can invoke services across domains. This functionality allows programmers to expand or partition applications without changing any application code.
The following figure shows the relationship between administrative commands and servers in the Domains administrative subsystem.
Figure 4-1 Domains Run-Time Administration
The BEA Tuxedo Domains component offers the following administrative commands:
BDMCONFIG
) while the BEA Tuxedo application is running. The command acts as a front-end process that translates administrative commands and sends service requests to the DMADMIN
service, a generic administrative service advertised by the DMADM
server. The DMADMIN
service invokes the validation, retrieval, or update of functions provided in the DMADM
server to maintain the BDMCONFIG
file.dmadmin
command. It provides a registration service to domain gateway groups. This registration service is requested by GWADM
servers as part of their initialization procedure. The registration service downloads the configuration information required by the requesting domain gateway group. The DMADM
server maintains a list of registered domain gateway groups, and propagates to these groups any changes made to the configuration.DMADM
server to obtain the configuration information used by the corresponding domain gateway group. The GWADM
accepts queries from DMADM
to obtain run-time statistics or to change the run-time options of the corresponding domain gateway group. Periodically, the GWADM
server sends an "I-am-alive" message to the DMADM
server. If no reply is received from the DMADM
server, the GWADM
server registers again. This mechanism makes sure the GWADM
server always has the latest copy of the Domains configuration for its group.Note: For information about domain gateway types other than GWTDOMAIN
, see BEA eLink Documentation .
BDMCONFIG
—the binary version of the Domains configuration file, which together with the TUXCONFIG
file and factory_finder.ini
file (CORBA only) contain all the configuration parameters that the BEA Tuxedo software needs to create a Domains configuration.Note: You can also specify gateway parameters when a domain gateway group is booted using the CLOPT
parameter, when the GWADM
server is defined in the SERVERS
section of the TUXCONFIG
file.
dmadmin
is an administrative interface to the DMADM
and GWADM
servers. The communication between the two servers is done via FML typed buffers. Administrators can use the dmadmin
command in the following ways:
BDMCONFIG
file and the different domain gateway groups running within a particular BEA Tuxedo application.BDMCONFIG
file.Note: You can delete information from the BDMCONFIG
file at run time only if the deletions do not involve an active domain gateway group.
The Domains administrative server, DMADM(5), is a BEA Tuxedo-supplied server that performs the following functions:
BDMCONFIG
fileBDMCONFIG
fileThe DMADM
server advertises two services:
DMADMIN
, which is used by the dmadmin
command and the GWADM
server.DMADM_
svrid
, where srvid
is the appropriate server ID for the service. Registered GWADM
servers use DMADM_
svrid
for specific administrative functions (for example, to refresh the domain gateway group configuration information or to signal that a GWADM
is still registered).The DMADM
server must be defined in the SERVERS
section of the TUXCONFIG
file as a server running within a group (for example, DMADMGRP
). There should be only one instance of the DMADM
server in this group.
The gateway administrative server, GWADM(5), is a BEA Tuxedo-supplied server that provides administrative functions for a domain gateway group. The main functions of the GWADM
server include:
DMADM
server, and accepting queries from dmadmin
. The GWADM
server gets the domain gateway group configuration information by registering with the DMADM
server. The GWADM
server then makes the configuration available to gateways by storing the information in shared memory.dmadmin
for run-time statistics or by changing the run-time parameters of the domain gateway group.GWADM
server determines which transactions need to be logged by reading information stored in shared memory. When the GWADM
server is booted, it scans the log to see whether any transactions need to be recovered, and then reconstructs the transaction information in shared memory. The gateway server scans the information in shared memory and performs recovery for the corresponding transactions. The recovery procedure is performed asynchronously with new incoming or outgoing requests received by the domain gateway group.The GWADM
server advertises a service name based on the local domain access point name (as specified in the DM_LOCAL
section of the BDMCONFIG
file) associated with the domain gateway group to which the GWADM
server belongs. The dmadmin
command uses this service to retrieve information from all active domain gateway groups or from a specific domain gateway group.
The GWADM
server must be defined in the SERVERS
section of the TUXCONFIG
file. It should not be part of the MSSQ
used by the gateways associated with the group. It must be the first server booted within the domain gateway group; that is, either (a) it must have a SEQUENCE
number, or (b) it must be defined ahead of the gateway servers.
The GWADM
server requires the existence of a DMADM
server. Specifically, a DMADM
server must be booted before that GWADM
is booted.
The GWADM
server must create the shared memory required by the domain gateway group to populate the configuration tables with information received from the DMADM
server. The GWADM
server uses IPC_PRIVATE
with shmget
and stores the ipckey
returned in the shmid
field of its registry entry in the bulletin board. Gateways can obtain the ipckey
by retrieving the GWADM
registry entry and checking the shmid
field.
A domain gateway server provides connectivity to remote domain gateway servers, and can communicate with one or more remote gateways simultaneously. A gateway advertises the services imported to a BEA Tuxedo application and controls access to the local services exported by the application. You define your application's exported and imported services in the Domains configuration file (DMCONFIG
). Use dmadmin
to dynamically configure, monitor, and tune domain gateway groups.
BEA Tuxedo 9.x improves the GWTDOMAIN gateway performance while also keeping compatibility with other types of /Domain gateways. It limits most of the performance to the threaded platforms. It also allows other types of /Domain gateways to take advantage of this feature in the enhanced Common Gateway Architecture with simple program changes.
Many factors may affect performance of applications across multiple domains. For example:
Therefore, in order to observe achieved Domain Gateway performance improvement, the application has to minimize the above-mentioned factors. Otherwise the performance improvement for the gateway may not be significant.
Some suggestions for performance testing this feature are listed below:
Note: As a prerequisite, service processing time on the server side should not be time-consuming. The total response time contains both gateway processing time and service processing time. If service processing time is quite long, the performance improvement for the gateway is submerged.
Application programmers can request the execution of remote services within a transaction. Also, users of remote domains can request local services to be executed within a transaction. Domains, therefore, coordinates the mapping of remote transactions to local transactions, and the sane termination (commitment or rollback) of these transactions.
The BEA Tuxedo system architecture uses a separate process, the Transaction Manager Server (TMS
), to coordinate the commitment and recovery of transaction branches accessing a particular group. In a Domains environment, however, this architecture would require extra messages from the gateway to the TMS
server to process a commitment for an incoming transaction. To simplify the Domains architecture and to reduce the number of messages, the TMS
code is integrated with the gateway code. Thus, domain gateways can process the transaction protocol used by the BEA Tuxedo system. The BEA Tuxedo transaction protocol requires that the domain gateway group advertise the TMS
service, which is done when the first gateway is booted. Once the TMS
service is advertised, any transaction control messages directed to the domain gateway group are placed on the gateway's queue.
Domain gateway groups should be defined in the TUXCONFIG
file without the TMSNAME
, TMSCOUNT
, OPENINFO
, and CLOSEINFO
parameters. These four parameters apply only to groups that use an XA-compliant resource manager, which domain gateways do not use.
The commitment protocol across domains is strictly hierarchical. It is not possible to flatten the transaction tree because the structure of the transaction tree is not fully known by every domain; a superior knows only its immediately subordinate domains. Flattening the tree would also require the root domain to be fully connected to all domains participating in the transaction.
Domain gateways provide four capabilities that you can use to manage transactions. These capabilities are described in the following sections:
In the BEA Tuxedo system, the TMS
is a special server that is implicitly associated with server groups that use X/Open XA-compliant resource managers. The TMS
server releases application servers from the delays associated with the distributed 2-phase commitment protocol. TMS
servers coordinate the commitment of a transaction via special service requests to the TMS
service, which is offered by all TMS
servers.
In a Domains environment, GWTDOMAIN
gateways are not associated with an XA-compliant resource manager. The Transaction Processing Working Group (TPWG
) of X/Open has proposed an advanced XA interface. This interface is not used in the BEA Tuxedo system because the interface does not match the highly asynchronous and non-blocking model required by the gateway. While domain gateways do not use a separate TMS
server, they do offer the Transaction Manager Servers capability, which allows gateways to coordinate the 2-phase commitment of transactions executed across domains.
Domain gateways coordinate transactions across domains in the following manner:
TMS
service and perform all operations associated with that service. Messages sent to this service are placed on the queue used by the appropriate domain gateway group, and the gateways manage the transactions associated with the group.Figure 4-2 The Domain Gateway as Subordinate/Coordinator of Another Domain Gateway Group
Figure 4-3 Client Commit Managed by a Domain Gateway
AUTOTRAN
capability. When this combination is used, the last server in the forward chain (the domain gateway) issues the commit and becomes the coordinator of the transaction. (A domain gateway always acts as the last server in a forward chain.)AUTOTRAN
capability. This capability is required when an the application administrator wants to enforce reliable network communication with remote services. Administrators can specify this capability by setting the AUTOTRAN
parameter to Y
in the corresponding remote service definition. In the BEA Tuxedo system, a transaction tree is a 2-level tree where the root is the domain gateway group coordinating a global transaction and branches are involved in the transaction. Each group performs its part of the global transaction independently from the parts performed by other groups. Each group, therefore, implicitly defines a transaction branch. The BEA Tuxedo system, through TMS
servers, coordinates the completion of the global transaction, making sure each branch is completed.
A GTRID
is a G
lobal T
ransaction Id
entifier. GTRID
mapping defines how to construct a transaction tree that crosses domain boundaries. You specify GTRIDs
using the MAXGTT
parameter in the RESOURCES
section of the BEA Tuxedo configuration file.
In the X/Open DTP Model, a Transaction Manager Server can construct transaction trees by defining either tightly-coupled or loosely-coupled relationships with a resource manager (RM) by the way it interprets the transaction identifiers (XIDs
) used by the XA interface.
A tightly-coupled relationship is one in which a single transaction identifier, XID
, is used by all processes participating in a single global transaction, accessing a single RM. This relationship maximizes data sharing between processes; XA-compliant RMs expect to share locks for resources used by processes having the same XID
. The BEA Tuxedo system achieves the tightly-coupled relationship via the group concept; that is, all work done by a group on behalf of a given global transaction belongs to the same transaction branch; all the processes executed by the group are given the same XID
.
In a loosely-coupled relationship, the TMS
generates a transaction branch for each part of the work in support of the global transaction. The RM handles each transaction branch separately; there is no sharing of data or of locks between the transaction branches. Deadlocks between transaction branches can occur and result in the rollback of a global transaction. In the BEA Tuxedo application, when different groups participate in a single global transaction, each group defines a separate transaction branch, which results in a loosely-coupled relationship.
There are several differences between global transactions in a single BEA Tuxedo application and global transactions across domains. The first difference is that in the Domains framework, the transaction tree cannot be flattened to a 2-level tree. There are two reasons for this:
This means that the commitment protocol across domains must be hierarchical. Even a loopback service request defines a new branch in the transaction tree.
Note: A loopback request goes to another domain and then comes back to be processed in the original domain. For example, Domain A requests a service of Domain B. The service in Domain B requests another service in Domain A. The transaction tree has two branches at the network level: a branch b1 from A to B and a branch b2 from B to A. Domain A cannot commit the work done on branch b2 before receiving commit instructions from B.
The structure of a transaction tree for global transactions across domains also depends on the distributed transaction processing protocol used by a relevant domain gateway instantiation. For example, in the OSI TP protocol each dialogue (the OSI TP word for a service request) is associated with a different transaction branch. In the BEA Tuxedo system, the OSI TP instantiation uses a dialogue for each service request, so each service request is mapped to a separate transaction branch. The XAP-TP interface hides this mapping and provides a mechanism by which an entire OSI TP subtree can be referenced by a user-defined identifier. (In the BEA Tuxedo implementation, this identifier is the GTRID
.) The GTRID
is used to instruct XAP-TP how a transaction tree must be constructed, that is, which dialogues must be included within a given OSI TP transaction. Therefore, from the BEA Tuxedo perspective, a whole OSI TP subtree can be managed as a single transaction branch.
This property, however, applies only to outgoing service requests (that is, service requests sent from the root domain to subordinate domains). It cannot be applied to incoming service requests. The OSI TP instantiation consequently implements a loosely-coupled relationship; each incoming service request is mapped to a new BEA Tuxedo global transaction.
The TDomain instantiation tries to optimize GTRID
mapping by implementing a tightly-coupled relationship. In TDomain, multiple service requests issued on behalf of the same global transaction are mapped to the same network transaction branch. Therefore, incoming service requests can be mapped to a single BEA Tuxedo transaction. However, the hierarchical structure of interdomain communication and the interdomain transaction tree must still be maintained.
The optimization that TDomain introduces applies only to a single domain. When two or more domains are involved in a transaction, the network transaction tree contains at least one branch per domain interaction. Hence, across domains, the network transaction tree remains loosely-coupled. There are as many branches as there are domains involved in the transaction (even if all the branches access the same resource manager instance).
Domain gateway groups implement a loosely-coupled relationship because they generate different transaction branches for interdomain transactions.
The following figure shows the service request graph for a client that generates three service requests: one local request (r0
) and two remote requests (r2
and r3
). Request r0
goes to a local service (Svc0
), which generates another remote service request (r1
). Request r1
goes to remote service Rsvc1
, which issues a loopback service request r4
to local service Svc4
. Svc0
and Svc4
are executed in different groups (G0
and G4
). The domain gateway is executed within another group (GW
), and the remote services Rscv1
, Rsvc2
, and Rsvc3
are executed in another domain (Domain B).
Figure 4-4 Service Request Graph
The following two figures show the transaction tree for BEA eLink OSI TP and the transaction tree for BEA Tuxedo domains. It is assumed, in these figures, that both Domain A and Domain B are BEA Tuxedo system applications.
BEA eLink OSI TP is loosely-coupled because of the OSI TP protocol. The transaction tree for this instantiation shows group G0 in Domain A coordinating the global transaction started by the client. Group G0
coordinates group GW
. Requests r1
, r2
, and r4
are mapped each to an OSI TP dialogue and therefore to an OSI TP transaction branch. However, OSI TP uses the XAP-TP feature that allows an entire OSI TP transaction to be referred by a unique identifier (T1
) and uses this identifier for requests r1
, r2
, and r3
. It is up to XAP-TP to generate OSI TP transaction identifiers and to construct the corresponding OSI TP transaction tree. The only function that must be performed by the generic Domains software is the mapping of service requests r1, r2, and r3 to the T1
identifier.
In Domain B, OSI TP uses the rule that new transaction branches must be mapped to a new BEA Tuxedo transaction. Therefore, OSI TP transaction branches r1
, r2
, and r3
get mapped to three different BEA Tuxedo transactions (the corresponding mapping is represented by identifiers T2
, T3
, and T4
). The graph shows the domain gateway group GW
in Domain B coordinating three BEA Tuxedo transactions on group G1
.
Finally, there is the loopback service request r4
that generates another branch in the transaction tree. OSI TP maps this request to identifier T2
, but XAP-TP generates a new branch in its transaction tree (r4
: B to A'). This is a new transaction branch on Domain A, and therefore, the gateway generates a new mapping T5
to a new BEA Tuxedo transaction. Therefore, the transaction graph shows that domain gateway group GW on Domain A coordinates group G4
.
Notice that the hierarchical nature of the OSI TP protocol is fully enforced by these mappings. However, because these mappings introduce a loosely-coupled relationship, the probability of intratransaction deadlock is increased (for example, there are three BEA Tuxedo transactions accessing the RM represented by group G1
).
Figure 4-5 Transaction Tree for BEA eLink OSI TP Environment
The TDomain instantiation provides a tightly-coupled integration that solves this deadlock problem by minimizing the number of transaction branches required in the interoperation between two domains. The corresponding transaction tree is shown in the following figure.
Figure 4-6 Transaction Tree for TDomain Environment
Notice that the gateway still must perform mappings between a BEA Tuxedo system transaction and a network transaction, and that the hierarchical nature of the communication between domains must be strictly enforced. The diagram shows that requests r1
, r2
, and r3
are mapped to a single TDomain transaction branch. Therefore, on Domain B only one BEA Tuxedo system transaction needs to be generated; T2
represents this mapping and the graph shows domain gateway group GW
on Domain B coordinating group G1
. Request r4
is mapped to identifier T2
on Domain B, but TDomain will generate a new branch in its transaction tree (r4
: B to A'). Because this is a new transaction branch on Domain A, the gateway generates a new mapping, T3
, to a new BEA Tuxedo system transaction. The graph shows that domain gateway group GW
on Domain A also coordinates group G4
. Hence, the hierarchical nature of interdomain communication is fully enforced with this mapping: group G4
cannot commit before group G1
.
Domains transaction management can be summarized as follows:
Logging is used to keep track of the progress of a 2-phase commit protocol. The information stored in the log is used to make sure a transaction is completed in the event of a network failure or machine crash.
To ensure completion of transactions across domains, domain gateways log the mapping between local and remote identifiers. Along with this information, the Domains transaction management facility records the decisions made during different phases of the commitment protocol, and any information available about the remote domains involved in the transaction. In the OSI TP case, the XAP-TP interface logs the information required for the recovery of the OSI TP protocol machine. The information is referred to as a blob (binary large object) and is kept in the same log record as the commit information to make recovery easier.
Domains log records have a different structure from the log records stored in the BEA Tuxedo system TLOG
. TLOG
records are fixed in size and are stored in a single page. Domains log records vary in size; more than one page may be required to store the record. The Domains logging mechanism, DMTLOG
, has the capability of storing variable-size log records.
When a TMS
is the superior of a domain gateway group, the BEA Tuxedo TLOG
is still required to coordinate the commitment.
Logging is performed by the GWADM
administrative server. The request for a log write is made by the GWTDOMAIN
process, but the actual log write is performed by the GWADM
process.
You must create a log called DMTLOG
for each domain gateway group. The DMTLOG
files are defined in the DM_LOCAL
section of the DMCONFIG
file. To create a DMTLOG
file, add an entry for the DMTLOGDEV
parameter:
DMTLOGDEV=
string
where string
is the name of the log file. In addition, you cam set one or both of the two optional parameters:
For more information, see DMCONFIG(5) in BEA Tuxedo File Formats, Data Descriptions, MIBs, and System Processes Reference.
Administrators also have the option of using the run-time administration utility (dmadmin
) to create a DMTLOG
. For more information, see dmadmin(1) in BEA Tuxedo Command Reference.
If a DMTLOG
has not been created when a domain gateway group is booted, the gateway server automatically creates the log, based on information in the BDMCONFIG
file.
Until a logging device is specified in the BDMCONFIG
file, a domain gateway group cannot process requests in transaction mode and the domain gateway group cannot offer the TMS
service.
To coordinate the commit protocol, domain gateways require the following two log records:
When a transaction has been committed on all machines, these logs for the transaction are removed.
When the OSI TP protocol is being used, two types of heuristic records are logged:
Heuristic log records persist until they are explicitly removed by the administrator. This persistence is required to provide the correct information during recovery after a crash, and to provide diagnostic information for administrators.
The administrator uses the forgettran
command (run with tmadmin(1)) to remove heuristic records when they are no longer needed.
When a domain gateway group is booted, the gateway server performs an automatic warm-start of the DMTLOG
. The warm-start includes scanning the log to see if any transactions were not completed. If incomplete transactions are found, action is taken to complete them.
In OSI TP, any blobs stored in the DMTLOG
with a transaction record are passed to the network access module, which uses the blobs to reconstruct its internal state and to recover any failed connections
In the case of heuristic decisions, if a domain gateway group is a subordinate of a local TMS
and a heuristic decision has been indicated, the TMS
generates a TMS_STATUS
message to learn the final decision:
![]() |
![]() |
![]() |