| With the extensive application of internet, there is a growing need for information. As a main way to information analysis and processing, complex query of database is rather important. Thanks to efficient query, less storage space and fewer disk fetch, Column-Oriented database is more suitable to dispose and analyze massive data. Thus, Column-Oriented database is one of the most important directions in the field.Based on the protosystem(GSQL) implemented by the author’s team, the team has designed a set of simple and efficient vector query primitives for full usage of multi CPU architecture. The author has tested and analyze the GSQL system based on TPC-H and notices that fine grain primitive would lead to massive physical vector operations, which results in lowering the performance of selection operation. Thus, the paper put forward a optimal scheme: coarser-grained primitive and reuse intermediate results for runtime optimization.First, the author designs coarser-grained MULTI_SELECT primitives to implement multi-condition selection operation.Then puts forward that Reuse of intermediate results could optimize coarser-grained MULTI_SELECT primitives at runtime. The paper has implemented the reuse of intermediate results, which includes four parts: choice of intermediate results, reuse of intermediate results, management intermediate results, execution of primitives. At last, the author not only test the performance of reuse of intermediate results under Multi-conditional primitives but also discuss when to reuse intermediate results under different input, query order. Then, analyzes the influence to the reuse of intermediate results under caches of different sizes. Although the reuse of intermediate results perform well under multi-condition selection. However, the operation is running at cost of memory, which need users to make a balance between performance and memory. Thus, the management of intermediate results will become choke point to the performance of reuse of intermediate results. The paper only implements a better promotion of performance under the optimization of multi-condition selection which may offer reference value for the optimization of Column-Oriented Database. |
