Tandem Computers Incorporated

Compaq Computer Corporation, Inc.

Microsoft Technology Licensing, LLC

Hewlett-packard Development Company, L.p.

Microsoft Corporation

William J Carley

Sitaram V Lanka

Albert C Gondi

Johannes Klein

Michael John Skarpelos

Matthew C McCline

James Ernest Johnson

James M Lyon

Michael Ray Clark

John David Doty

Robbert Van Der Linden

Kapil Gupta

Yu-Cheung Cheung

Jonathan M Cargille

Charles Stuart Johnson

Max A Feingold

David J Wisler

Trina R Wisler

Robert Van Der Linden

James A Lyon

A transaction manager virtualization layer is provided for supporting multiple active transaction managers in a cluster or node. By virtualizing the transaction managers, applications and services are enabled to migrate across cluster nodes or even outside a cluster that includes relevant recovery mechanisms for transactions that are not completed at the time of migration. Multiple implementations of a transaction manager may be enabled to run side-by-side in a node or cluster allowing a rolling upgrade of transaction managers in clusters.

Transaction manager virtualization

The present invention extends to methods, systems, data structures, and computer program products for maintaining correct transaction results when transaction management configurations change. An intermediate transaction manager is transparently interposed between one or more resource managers and a plurality of transaction managers. The intermediate transaction manager multiplexes and intercepts calls and notifications between the one or more resource managers and the plurality of transaction managers. The intermediate transaction manager is configured to wrap recovery cookies with transaction manager type information and identity information that identify transaction managers. The information can be used to re-attach to the transaction manager to recover the correct outcome of the transaction. The intermediate transaction manager can also intercept recovery-complete calls directed to a specific transaction manager and send the recovery-complete call to each of the plurality of transaction managers. Resource managers can receive, store, and send wrapped recovery cookies.

Maintaining correct transaction results when transaction management configurations change

The present invention extends to methods, systems, and computer program products for synchronizing ambient state among multiple transaction managers. A coordinating transaction manager establishes transaction ambient state for an application initiated transaction. The coordinating transaction manager sends and a subordinate transaction manager receives a pointer to an ambient state delegate. The subordinate transaction manager sends a request to invoke the ambient state delegate to the coordinating transaction manager in response to receiving an operation. The coordinating transaction manager invokes the ambient state delegate to format the transaction ambient state for the subordinate transaction manager. The coordinating transaction manager sends and the subordinate transaction manager receives the formatted transaction ambient state such that the subordinate transaction manager can appropriately process the operation within the scope of the transaction. The subordinate transaction manager utilizes the formatted transaction ambient state to appropriately process the operation within the scope of the transaction.

Synchronizing transaction ambient state among multiple transaction managers

A Registration protocol is used in transaction processing for normal operations. If an error occurs, the system reverts to a Full Broadcast protocol. The Registration Protocol reduces the number of messages that are sent among CPUs in a cluster thereby permitting performance improvements in the system. The Registration Protocol has Begin, DPCheck, PhaseFlush and Phase(lock release) phases just as does the Full Broadcast Protocol, thereby permitting the Full Broadcast protocol to step in at any phase after an error is detected.

Hybrid method for flushing transaction state in a fault-tolerant clustered database

Methods and systems that provide a resource manager the ability to register with a transaction manager that it needs to be &ldquo;pre-prepared&rdquo; for a transaction. The transaction manager will issue a pre-prepare notification to any resource manager so registered, and will delay phase one of the two-phase commitment (2PC) protocol until all pre-prepare-registered resource managers have responded that they are ready for commitment processing.

Volatile resource manager with pre-prepare notification

Incorporation of a collateral process as a participant in a transaction is made possible by a method and system in accordance with the present invention. Typically, after the transaction is initiated, the collateral process is called and then is registered as a participant,in the transaction. A prepare signal is sent to each registered collateral process when end stage of the transaction is reached. Then, a ready signal is received from the collateral process if the collateral process is completed successfully; and an abort signal is received from the collateral process if the collateral process does not complete successfully or a violation is detected. If a ready signal is received, a commit record is written to a log, and a commit signal is sent to each registered collateral process. In response to the commit signal, a forgotten signal is received from each registered collateral process.

Incorporating collateral and concurrent activity in a data processing transaction

A transaction processing system, including multiple processor units communicatively interconnected, manages information collection by employing a distributed transaction management facility to track and make consistent changes. When each transaction is started, a data structure is created that maintains information concerning the transaction. Included in the data structure is the identity of all processor units having resources involved in the transaction. Should a processor unit fail, and the transaction management facility is notified of that failure, the data structures of all pending transactions will be examined to see if the failed processor unit had a resource that was a participant in the corresponding transaction. If so, the transaction management facility can then make a decision as to whether or not to abort the transaction.

Fault-tolerant transaction processing in a distributed system using explicit resource information for fault determination

Methods and systems for interoperating a transaction processing system implementing the presumed abort variation of the two-phase commit (2PC) protocol with a system implementing the presumed nothing variation. The gateway process that is responsible for exporting a transaction branch to a foreign transaction management system from a system implementing the presumed abort variation records whether the foreign transaction manager is implementing the presumed abort variation or the presumed nothing variation. After system failure, the gateway process determines which variation is adopted for a specific transaction branch. If the foreign transaction manager is implementing the presumed abort variation, the transaction manager from which the transaction branch originated communicates with the foreign transaction manager through the gateway process in accordance with the presumed abort variation. If the foreign transaction manager is implementing the presumed nothing variation, the gateway process takes charge of driving abort processing in a manner transparent to the originating transaction manager.

Logging of transaction branch information for implementing presumed nothing and other protocols

A fault tolerant computer system distributes audit trail files containing audit records, across an arbitrary number of disk volumes. After one audit trail file becomes full, audit records are directed toward a next audit trail file stored on a different disk volume. Storage of newly generated audit rotates through the disk volumes in round-robin fashion. Full audit trail files are eventually archived and their space becomes available again for renaming and storage of newly generated audit records. The number of audit records available for on-line recovery after a failure is not limited to the storage capacity of any single disk volume. Furthermore, there is no contention for disk access between archiving of full audit trail files and storage of newly generated audit records.

Multi-volume audit trails for fault tolerant computers

System-critical files are protected from being inadvertently modified or deleted by placing them in a Reserve name space of storage that requires any process seeking modifying access to the Reserve name space to have a 'Right' to do so. The right to modifying access is garnered by a process first making a call to a system library procedure that causes a memory-stored data structure associated with the calling process to be modified, identifying the process as one with a Right to make modifying access to the Reserve name space. Any attempt to modify, delete, or create any file residing in the Reserve name space without the Right will be refused.

Method and apparatus for preventing inadvertent changes to

A computing system comprising a number of multiple processor unit nodes interconnectively connected by a communications system to form a cluster is initialized in a manner that creates separate, independent execution environments for each processor unit. Each multiple processor unit node is configured to operate as a symmetric multiprocessing system with a single, shared memory. During a Startup procedure, memory area segments are created for each processor unit, providing processes running on that processor unit with mutually exclusive access to the associated memory area segment. Startup determines an order of the nodes of the cluster system, and establishes a Coordinator process in the lowest numbered processor unit in the first node of the order. The Coordinator process, when created, directs the remainder of the Startup procedure, and constructs a succession list that identifies the next processor unit in the order to take over the Coordinator process should the first fail.

Method for sequential and consistent startup and/or reload of multiple

A fault tolerant computer system distributes audit trail files containing audit records, across an arbitrary number of disk volumes. After one audit trail file becomes full, audit records are directed toward a next audit trail file stored on a different disk volume. Storage of newly generated audit rotates through the disk volumes in roundrobin fashion. Full audit trail files are eventually archived and their space becomes available again for renaming and storage of newly generated audit records. The number of audit records available for on-line recovery after a failure is not limited to the storage capacity of any single disk volume. Furthermore, there is no contention for disk access between archiving of full audit trail files and storage of newly generated audit records.

Growing community of inventors

San Jose, CA, United States of America

William J Carley