Optimized system for analytics (graphs and sparse matrices) operations

摘要

A graph processing system includes a graph API (Application Program Interface), as executed on a processor of a computer and as capable of implementing any of a plurality of graph operators to express computations of input graph analytics applications. A run-time system, executed by the processor, implements graph operators specified by each graph API function and deploys the implemented graph operators to a selected computing system. A library contains multiple implementations for each graph API function, each implementation predetermined as being optimal for a specific set of conditions met by a graph being processed, for functional capabilities of a specific computing system on which the graph is being processed, and for resources available on that specific computing system.

主权项

1. A method, comprising:
storing, in a library accessible to a processor on a computer, a set of multiple implementations of graph analytics operators for executing a plurality graph operations, each implementation predetermined as optimal for a specific set of conditions met by a graph being processed, a computing system's functional capabilities on a potential computing system upon which the graph is to be processed, and resources available on that computing system; receiving, through a graph processing interface (GPI) being executed by the processor, a request and information for an input graph analytics application to be executed; receiving, through a resource application program interface (API) being executed by the processor, information on computation capabilities of computing systems potentially capable of performing operations involved in executing the input graph analytics application; selecting which graph analytics operator implementation from the library and which potential computing system would be optimal to execute the operations on input data for the input graph analytics application; deploying graph data from the input graph analytics application to the selected computing system for execution; and receiving computation results from the selected computing system and forwarding the computation results back through the GPI to satisfy the input graph analytics application request.