Abstract: A method for constructing a projection for storing data on a storage device may include causing a processor to extract characteristics of expected output of a query on the data, and to construct a projection for the data based on the extracted characteristics.

Abstract: A system and method for switching algorithms during a run time computation, the method including configuring hardware of a networked cluster of processing elements, each processing element with a memory hierarchy, to perform a first-tier algorithm on input data, the input data having cardinality and stored on one or a plurality of nodes in the networked cluster. Performing at least a portion of a second-tier algorithm and determining whether to complete the second-tier algorithm and perform a third or subsequent tier algorithm, the determination dependent on cardinality. Automatically passing data to an output if the cardinality of the second-tier algorithm is greater than a threshold cardinality, and passing the data back to the second-tier algorithm or to one or a plurality of subsequent algorithms, in response to the cardinality being less than the threshold, and automatically passing the data to an output at the completion of the data processing.

Abstract: A method of analyzing an expression to infer output data properties, comprises storing, in a storage device, a root node “e” in R, in which “e” is an expression and R is a root node of an abstract syntax tree (AST) of the expression “e”, storing a number of child nodes of R into C1, . . . Cn, with a processor coupled to the storage device, calling analyze(Ci), storing output properties of expression C into Pi I, and determining if any of the child nodes Ci in {C1 . . . Cn} has not been analyzed, in which if the last child node has been analyzed, with the processor, calling compute-output-properties(R, P1, . . . Pn), storing, in the storage device, the result in Poutput, and returning Poutput as the output properties of the expression “e.