Oracle International Corporation

Roger J Bamford

Boris Klots

Sashikanth Chandrasekaran

David Brower

Neil MacNaughton

Vinay H Srihari

Wilson Wai Shun Chan

William Bridge

Angelo Pruscino

Garret Frederick Swart

Amit Ganesh

William Havinden Bridge, Jr

Neil Macnaughton

Juan R Loaiza

Jonathan D Klein

Alok Kumar Srivastava

Gary C Ngai

Vikram Joshi

J William Lee

Srinivas Vemuri

Wilson Wai Chan

Arvind Nithrakashyap

Xiang Liu

James William Stamos

Lawrence W Jacobs

Wei Huang

Alan J Demers

Alok Pareek

Tak Fung Wang

Wei Ming Hu

Leng Leng Tan

Brian M Oki

Jeffrey Fischer

Bhagat V Nainani

Zheng Zeng

Gary Hallmark

Shehzaad Nakhoda

Sandeep Jain

Kenneth R Jacobs

Ravi Prem Mirchandaney

Tudor Bosman

Robert N Miner

Dirk A Kabcenell

Forrest W Howard

William Jr Bridge

Srinivas S Vemuri

Techniques are provided for managing caches in a system with multiple caches that may contain different copies of the same data item. Specifically, techniques are provided for coordinating the write-to-disk operations performed on such data items to ensure that older versions of the data item are not written over newer versions, and to reduce the amount of processing required to recover after a failure. Various approaches are provided in which a master is used to coordinate with the multiple caches to cause a data item to be written to persistent storage. Techniques are also provided for transferring data items and locks associated with the data items from one node to another.

Methods to perform disk writes in a distributed shared disk system needing consistency across failures

Various techniques are described for improving the performance of a shared-nothing database system in which at least two of the nodes that are running the shared-nothing database system have shared access to a disk. Specifically, techniques are provided for recovering the data owned by a failed node using multiple recovery nodes operating in parallel. The data owned by a failed node is reassigned to recovery nodes that have access to the shared disk on which the data resides. The recovery logs of the failed node are read by the recovery nodes, or by a coordinator process that distributes the recovery tasks to the recovery nodes.

Parallel recovery by non-failed nodes

Techniques are provided for lock management. The techniques are based on an enhanced lock management system that generates a semantic response in response to lock requests for a resource. The semantic response communicates both the underlying cause blocking the request, and information that may be used by the requester to obtain notification of when the underlying cause should no longer lead to denial of the lock request. The semantic response may be generated by the master of the resource, who provides the semantic response to the local lock manager of the lock requester. The semantic response may be retained by the local lock manager so that the semantic response can be provided to subsequent lock requesters, without need for interacting with another lock manager on another node.

Semantic response to lock requests to reduce coherence overhead in multi-node systems

Techniques are provided for managing caches in a system with multiple caches that may contain different copies of the same data item. Specifically, techniques are provided for coordinating the write-to-disk operations performed on such data items to ensure that older versions of the data item are not written over newer versions, and to reduce the amount of processing required to recover after a failure. Various approaches are provided in which a master is used to coordinate with the multiple caches to cause a data item to be written to persistent storage. Techniques are also provided for managing checkpoints associated with the caches, where the checkpoints are used to determine the position at which to begin processing recovery logs in the event of a failure.

Managing checkpoint queues in a multiple node system

Techniques are provided for determining which data item version to supply to a query. According to the techniques, the determination is made by associating a new field, which indicates the time a data item version was current, with each data item version; associating a new field with each query, which indicates the last change that the query must see made by the transaction to which the query belongs; and determining which data item version to use to answer the query based, in part, on a comparison between the values of the two new fields.

Consistent read in a distributed database environment

Various techniques are described for improving the performance of a shared-nothing database system in which at least two of the nodes that are running the shared-nothing database system have shared access to a disk. Specifically, techniques are provided for changing the ownership of data in a shared-nothing database dynamically, based on factors such as which node would be the most efficient owner relative to the performance of a particular operation. Once determined, the ownership of the data may be changed permanently to the new owner, or temporarily for the duration of the particular operation.

Dynamic reassignment of data ownership

A method and mechanism is described for reliably transmitting messages over an unreliable transport mechanism. A sender transmits a first message over an unreliable mechanism to a receiver, and a second message that identifies the first is transported over a reliable transport mechanism to the receiver. When the receiver receives the second message, the receiver determines whether it has received the first message. If not, the receiver requests retransmission of the first message.

Transmitting data reliably and efficiently

Various techniques are described for improving the performance of a shared-nothing database system in which at least two of the nodes that are running the shared-nothing database system have shared access to a disk. Specifically, techniques are provided for changing the ownership of data in a shared-nothing database without changing the location of the data on persistent storage. Because the persistent storage location for the data is not changed during a transfer of ownership of the data, ownership can be transferred more freely and with less of a performance penalty than would otherwise be incurred by a physical relocation of the data. Various techniques are also described for providing fast run-time reassignment of ownership. Because the reassignment can be performed during run-time, the shared-nothing system does not have to be taken offline to perform the reassignment. Further, the techniques describe how the reassignment can be performed with relatively fine granularity, avoiding the need to perform bulk reassignment of large amounts of data across all nodes merely to reassign ownership of a few data items on one of the nodes.

Ownership reassignment in a shared-nothing database system

Various techniques are described for improving the performance of a multiple node system by allocating, in two or more nodes of the system, partitions of a shared cache. A mapping is established between the data items managed by the system, and the various partitions of the shared cache. When a node requires a data item, the node first determines which partition of the shared cache corresponds to the required data item. If the data item does not currently reside in the corresponding partition, the data item is loaded into the corresponding partition even if the partition does not reside on the same node that requires the data item. The node then reads the data item from the corresponding partition of the shared cache.

Maintaining a shared cache that has partitions allocated among multiple nodes and a data-to-partition mapping

A method and apparatus are provided for improving the performance associated with transferring a data item and obtaining a lock in a multi-node system by interpreting the block transfer message as a lock grant message. Typically when a Holder of a data item transfers a data item (e.g. block) to a Requestor of that data item, the Holder will down-convert its lock for that data item and send a message to the Master of this data item indicating that a down-convert has been performed. Subsequently, the Master sends a lock grant message to the Requestor of the data item to inform the Requestor that it has been granted a lock for the data item. By interpreting the block transfer message as a lock grant message, the down-convert message and the lock grant message can be eliminated, which results in improved performance.

Method and apparatus for reducing latency and message traffic during data and lock transfer in a multi-node system

Growing community of inventors

Woodside, CA, United States of America

Roger J Bamford