rasdaman newsletter 09/2026
Scientific Computing Meets Datacubes
Bremen. Two Bachelor theses done at Constructor University show how rasdaman can run advanced scientific library functions directly inside an Array DBMS. By integrating Intel MKL, OpenCV and NumPy the server reduces costly data movement and exploits CPU capabilities optimally. This opens up new opportunities for climate data, satellite imagery, medical imaging and large-scale numerical analysis.
Supervised by Peter Baumann, Professor of Computer Science, the all-female team of Dulguun Ulziibayar and Zaara Valladares has integrated Intel MKL, OpenCV and NumPy into the rasdaman Array DBMS.
Callable directly through datacube queries, these functions keep computation close to the data, avoiding repeated export and re-import. Additionally, Intel MKL comes with particularly optimized code exploiting all CPU capabilities for highly efficient Big Data processing.
Ulziibayar implemented 15 functions across the three libraries, using three integration patterns and validating them with regression tests and an OpenCV visual demonstration. Valladares evaluated the integrated functions for correctness and performance, testing them across collections of varying dimensions; her results confirmed reliable, accurate output and showed that, while UDFs add a small overhead for some inputs, execution time improves as data volumes grow, making the approach scalable.
"This work illustrates how rasdaman’s UDF mechanism eliminates this overhead and allows the use of function calls from these libraries to occur through the use of rasql queries, while keeping computation server-side," says Dulguun Ulziibayar.
The theses show the integration is feasible and effective: tested functions produced accurate results across data types and dimensions, and UDF-based execution gets even more competitive as array sizes increase. By keeping processing inside the database, it reduces data-movement overhead and lays foundation for more scalable scientific analysis through humans and AI.
contact: Prof. Dr. Peter Baumann