Abstract: The present invention extends to methods, systems, and computer program products for executed stored procedures at parallel databases. Stored procedures are transformed so that execution of the stored procedure is split between a standalone database server and a parallel database coordinator. Execution of the stored procedure is initiated at the standalone database server. At execution time, control-flow statements, variable assignment, expression evaluation, etc., are handled by the standalone database server. SQL statements are passed from the standalone database server to the database for the execution. Results from executed SQL statements can be returned to the standalone database server or to a client. The parallel database coordinator can be added as a linked server to the standalone database server. In some embodiments, a session token is used to share session state between different parties.

Abstract: The present invention extends to methods, systems, and computer program products for optimizing queries of parallel databases. Queries can be partially optimized at an optimizer that is unaware of its use to optimize queries for parallel processing. The optimizer can produce a data structure (e.g., a SQL Server MEMO) that encapsulates a logical serial plan search space. The logical serial plan search space may not incorporate any notion of parallelism into the plan space itself. A parallel-aware optimizer can parallelize the logical serial plan search space by augmenting the data structure (e.g., transforming the SQL Server MEMO into a parallel MEMO). Augmentation can be with data movement operations that move data associated one or more compute nodes in a distributed architecture. Cost estimates can be calculated for the operations contained in the parallelized data structure. The parallel plan with the lowest estimated cost can be selected for the query.

Abstract: The present invention extends to methods, systems, and computer program products for formulating global statistics for parallel databases. In general, embodiments of the invention merge (combine) information in multiple compute node level histograms to create a global histogram for a table that is distributed across a number of compute nodes. Merging can include aligning histogram step boundaries across the compute node histograms. Merging can include aggregating histogram step-level information, such as, for example, equality rows and average range rows (or alternately equality rows, range rows, and distinct range rows), across the compute node histograms into a single global step. Merging can account for distinct values that do not appear at one or more compute nodes as well as distinct values that are counted at multiple compute nodes. A resulting global histogram can be coalesced to reduce the step count.

Abstract: The present invention extends to methods, systems, and computer program products for performing parallel joins on distributed database data. Embodiments of the invention include a phased semi-join reduction strategy using replication and shuffle operations to join a first and a second data source. A filter building phase uses replication and pushes down a “Distinct” (e.g., SQL) operator to produce a list of join keys for the first data source (one side of the join). A shuffle phase for the second data source is modified to join to the key list produced in the first phase as a row filtering mechanism. A join phase then joins the first and second data sources.

Abstract: The present invention extends to methods, systems, and computer program products for performing parallel joins on distributed database data. Embodiments of the invention include a phased semi-join reduction strategy using replication and shuffle operations to join a first and a second data source. A filter building phase uses replication and pushes down a “Distinct” (e.g., SQL) operator to produce a list of join keys for the first data source (one side of the join). A shuffle phase for the second data source is modified to join to the key list produced in the first phase as a row filtering mechanism. A join phase then joins the first and second data sources.

Abstract: Techniques and systems are disclosed for returning temporally-aware results from an Internet-based search query. To determine if a query is temporally-based one or more query features are collected and input into a trained classifier, yielding a temporal classification for the query. Further, if a query is classified as temporal, the query results are shifted by determining an alternate set of results for the query, and returning one or more alternate results to one or more users. Based on user interactions with the one or more alternate results, the classifier can be updated, for example, by changing the query to a non-temporal query if the user interactions identify it as such.

Abstract: Techniques and systems are disclosed for returning temporally-aware results from an Internet-based search query. To determine if a query is temporally-based one or more query features are collected and input into a trained classifier, yielding a temporal classification for the query. Further, if a query is classified as temporal, the query results are shifted by determining an alternate set of results for the query, and returning one or more alternate results to one or more users. Based on user interactions with the one or more alternate results, the classifier can be updated, for example, by changing the query to a non-temporal query if the user interactions identify it as such.