Graph query expenditure method for complete historic records

Abstract

The invention provides a graph query expenditure method for complete historic records. The method comprises the following steps: Step 1, during loading of original data, servers collect data required for expenditure predictions and carry out statistics of the data; Step 2, each server receives a query request from a client and analyzes query statements in the query request; Step 3, triple query orders are selected, and whether the query orders are reasonable or not is judged; and Step 4, query expenditures are predicted. The method provided by the invention is used for graph query expenditure predictions for the complete historic records, and has the advantages that characteristics of a complete historic record query mode is fully utilized, the query expenditures are accurately estimated, the efficient query orders are selected, and characteristics of a high-performance network are fully utilized; effects of communication of the high-performance network in a distributed environment on query delays are further considered; and a delay of a specific query request can be predicted according to a model, and the triple query order with the lowest expenditures is selected, so that the overall performance of a graph query system is improved.

Description

technical field [0001] The invention relates to a method for predicting graph query overhead, in particular to a graph query overhead method for complete historical records. Background technique [0002] In the era of big data, information presents a highly unstructured, free and rich correlation. The social network graphs of many knowledge bases such as Weibo and Facebook are usually stored in the form of RDF (Resource Description Framework). Online query requests for graph data become more complex and free at the same time. Predicting and modeling the query overhead, and reasonably arranging the execution order of query statements according to the overhead is one of the important methods to reduce the delay of query statements. Correct and reasonable overhead methods can minimize useless intermediate results and make full use of High-performance network and hardware resources accelerate user query requests. [0003] SPARQL is a query language for "rich" semantic retrieva...

AI Technical Summary

Problems solved by technology

[0004] Existing overhead methods for graph query often consider single-step pruning for cost prediction, which cannot be applied to more efficient full-history pruning methods in high-performance networks, often resulting in inaccurate cost predictions, resulting in inaccurate Reasonable query sequence brings large query delay

In addition, the existing overhead methods often ignore the estimation of the communication overhead and the final merge operation overhead. The overhead of these two parts often accounts for a large proportion of the overall overhead of the query. Ignoring these two overheads will cause the overhead method to be inaccurate. , which leads to a large deviation from the true overhead

Images