Technology Blogs by SAP
Learn how to extend and personalize SAP applications. Follow the SAP technology blog for insights into SAP BTP, ABAP, SAP Analytics Cloud, SAP HANA, and more.
cancel
Showing results for 
Search instead for 
Did you mean: 
Taesuk
Product and Topic Expert
Product and Topic Expert

In early 2023, I had written "Faster query execution using lesser memory in SAP HANA Cloud" to show improvements in the SQL execution engine with faster execution using lesser memory. Since then, further enhancements have been done to improve the overall performance with using even lesser memory with the latest release.

Based on workloads captured from SAP Analytics Cloud, we have conducted performance comparison tests between QRC 2/2023 and the latest QRC 1/2024 instances.

For comparisons, we have compared in 2 different modes 

  • Cold Mode : Refers to the initial execution when the execution plan has not been cached. The execution requires both plan compilation and HEX L cache optimization.
  • Warm Mode : Refers to the 4th execution where the execution plan have been compiled and cached during the 1st execution and then JIT L code is also optimally cached within HEX.

Additionally, we have split the workload in 3 categories in terms of execution time taken.

  • Short : Execution < 1s
  • Mid : 1 ~ 10s
  • Long : > 10 s

First, lets check the performance based on CPU as HEX is executing in data chunks and parallelized pipelines.

CPU.png

Based on the above, we can assume

  • for the short scenario cold mode, the total and geo-mean CPU time is taking a slightly longer as more HEX feature coverage is added and thus more time fore compilation and L cache optimization. 
  • whereas the warm mode execution for shot scenarios have reduced more than 20%
  • as the cpu execution takes longer such as mid and long scenarios, the total and geo-mean CPU time gets even shorter in both mid and long scenarios.

Lets now check based on peak memory

Memory.png

  • For cold mode in all scenarios, the average peak memory have been reduced as the compilation have been further optimized.
  • memory reduction is also seen in shorter executions for warm mode.

Conclusion

With the QRC 1/2024 release, we have further optimized HEX in terms of performance with lower peak memory usage while extending the coverage of execution. Please expect additional improvements in our SQL execution engine for 2024.

Tae Suk Son
SAP HANA Cloud Product Management