摘要:
Disclosed is an evaluation technique for text search with black-box scoring functions, where it is unnecessary for the evaluation engine to maintain details of the scoring function. Included is a description of a system for dealing with blackbox searching, proofs of correctness, as well experimental evidence showing that the performance of the technique is comparable in efficiency to those techniques used in custom-built engines.
摘要:
Database tables which are to be loaded with large data objects are defined with a special logging attribute that causes the associated database management system to suspend recovery logging for the databases so defined, but to continue with logging control information, such as space management changes. During a two-phase commit operation for a database transaction in a distributed system, at each database all modified data pages are written to disk before the end of the phase 1 commit, if the pages have not been previously written. This and other write operations are monitored for errors. If any of the changed data pages cannot be written to disk successfully before the end the phase 1 commit operation, then the transaction is rolled back at all databases.
摘要:
Database tables which are to be loaded with large data objects (LOBs) are defined with a special logging attribute value. This attribute value causes the database management system to suspend recovery logging for the databases so defined, but to continue with the logging of control information, such as space management changes. Certain control information recovery log entries have a special status indicator. For LOBs, this status indicator is set to cause the associated LOB entry to be marked as invalid when the log entry is reapplied, for example, during a recovery operation from a media failure. Rather than invalidating the entire table, the database management system will only prevent access to the entries which have been invalidated during the recovery process.
摘要:
The present invention provides a system and method for utilizing the proximity of keys in sequential or near sequential index probes to avoid complete index tree traversal. Page information from three pages (LAST, PARENT and NEXT) are stored in separate information fields within an Index Lookaside Buffer. The LAST information field contains information on the most recent leaf page accessed during an index probe in a read key or an insert key operation, the PARENT information field contains information on the parent page of the most recently accessed leaf page described in the LAST information field, and the NEXT information field contains information on the most recent leaf page accessed during a fetch-next key or delete key operation. On a subsequent index probe, the LAST, NEXT and PARENT information fields are sequentially analyzed to determine if the search key is contained within the leaf page described by the LAST or NEXT information fields or if the search key is contained within one of the leaf pages pointed to by the parent page described by the PARENT information field. If none of the LAST, NEXT or PARENT information fields identify the leaf page containing the search key then the default root-to-leaf traversal would commence.
摘要:
Embodiments provide an approach for efficiently searching a temporal key index as well as enforcing key distinctiveness property in a database table corresponding to the index. For example, by addressing time range overlaps in an index having a sort order for temporal keys, a database management tool of a server system may efficiently process client requests while enforcing a temporal key property in the database table.
摘要:
Database elements are inserted into a database object by processing each of a plurality of operations in a sequential order within a first processing round to insert the database elements into the database objects, where processing for at least one operation in the order becomes suspended due to a resource request, and where at least one successive operation is initiated in response to suspension of one or more prior operations to enable prefetching of information for processing the operations. Each suspended operation is re-processed with the prefetched information in one or more additional processing rounds until processing of the operations is completed.
摘要:
Provided are techniques for rejecting rows while locating a target row. For a row that is stored in a hash space, a row filter value is generated for that row, and the row filter value is stored with the row. While trying to locate the target row in a collision chain in the hash space, a row filter value is calculated for the target row. For the row in the collision chain, the stored row filter value of the row in the collision chain is compared with the computed row filter value of the target row. In response to determining that the stored row filter value does not match the computed row filter value, it is determined that the row in the collision chain is not the target row.
摘要:
Techniques are described for processing a query to produce query results, the query specifying at least a first timestamp value. Embodiments receive the query for processing and access a database index containing a plurality of database keys. The database index contains one or more database index keys, each of which includes at least a timestamp value and a time zone value. Embodiments compare the first timestamp value specified in the query with a portion of one the database index keys to locate at least a portion of the query results. More specifically, the compared portion of the database index key excludes the time zone value. The located portion of the query results is then retrieved.
摘要:
A system, method and computer readable medium containing programming structions for an improved latch mechanism for use in a programming environment running concurrent tasks is disclosed. According to one aspect of the present invention, a latch is provided to a first task which is requesting access to a resource. After such access, the first task holds the latch, which is also released from the first task. The first task holding the latch marks the latch stealable, and if the latch is not marked stolen by another task, the first task can mark the latch unstealable. In another embodiment of the present invention, if the first task is provided a latch marked stealable, the resource associated with the stealable latch will be placed in a consistent state before the first task accesses the resource.
摘要:
Methods and apparatus, including computer program products, implementing and using techniques for performing an index scan in a database system. A sequence of pages visited during an index scan are dynamically tracked. The tracking includes tracking the number of page accesses requiring a wait for input or output. It is determined whether prefetching of an object is desirable. The determination being made at least in part based on the number of page accesses requiring a wait for input or output. If prefetching is desirable, a prefetching method is selected among a set of available prefetching methods. An effectiveness of the selected prefetching method is continuously monitored by tracking the sequence of pages accessed and available access methods. A different prefetching method is selected from the set of available prefetching methods if the different prefetching method has a better effectiveness for a portion of the index that is being scanned.