TMF Operations and Recovery Guide (G06.24+)

Routine Maintenance

HP NonStop TMF Operations and Recovery Guide—522417-002

2-20

Maintaining Distributed Transactions

TMF can service two kinds of distributed transactions:

•

Homogeneous distributed transactions, where portions of a transaction are

processed by TMF systems on two or more Expand nodes (NonStop systems).

•

Heterogeneous distributed transactions, where portions of a transaction are

processed under the control of multiple transaction managers: TMF and one or

more foreign transaction managers.

For homogeneous distributed transactions, a communication link is managed by the

Expand or FOX subsystem that connects two Expand nodes in a computer network;

when the Expand or FOX subsystem ceases to function, the link goes down.

For heterogeneous distributed transactions, a communication link is the logical

connection of TMF with a gateway process, accomplished through use of a resource

manager. When the gateway process opens the resource manager file, the link comes

up; when the gateway process closes the resource manager file, the link goes down.

(There are three kinds of failure that can cause a link to go down: communication

subsystem failure, gateway process failure, or TMF failure.)

When a homogeneous transaction in your local TMF system requires access to a

table, file, or server on a remote TMF system, the TMP in the local system sends a

message to the TMP in the remote system, which then creates and manages

information about the transaction (in essence, the transaction now exists in two

separate systems, and is managed cooperatively by two different TMPs).

When a heterogeneous transaction in your local TMF system requires access to a

table or file in a database managed by a foreign transaction manager, TMF and the

foreign transaction manager cooperate in achieving this access.

The system in which a distributed transaction originates is referred to as the parent

node while the secondary system, whose resources the transaction requires, is

referred to as the child node. Thus, when a node sends a transaction to another node,

the first (sending) node is the parent node, and the second (receiving) node is the child

node. If the second node requires resources from another system, the second node is

now both a child to the first node, and a parent to the third node; a node can have

more than one child node for a given transaction.

Maintaining distributed transactions requires an understanding of their normal

processing states and their dependence on the availability of communication links.

Note. Heterogeneous transactions, and the resource managers that support them, are used in

HP products such as HP NonStop TUXEDO. System managers working with them must

understand many issues involving multiple software subsystems and inter-platform

considerations; complete discussion of heterogeneous transactions lies beyond the scope of

this manual.