MemArk
13 min readAug 18, 2021

--

Figure 1. Typical AI driven online real-time decision system architecture
Figure 2. Example of credit card anti-fraud features
Figure 3. Performance of mainstream commercial databases on real-time feature extraction workload
Figure 4. Structure diagram of machine learning database OpenMLDB
Figure 5. Example of multi-window feature extraction by FEQL
Figure 6. The two-layer skiplist table structure of OpenMLDB storage engine
Figure 7. Performance comparison between DRAM version of OpenMLDB and other commercial databases
Figure 8. OpenMLDB using different persistent memory modes
Figure 9. Inconsistency of compare and swap (CAS) operation on persistent memory
Figure 10. Persistent CAS (PCAs) and smart pointers
Figure 11. Persistent hop table structure based on smart pointer
Figure 12. Database abbreviations and system configuration used in the experiment
Figure 13. Performance comparison of OpenMLDB based on DRAM and persistent memory
Figure 14. OpenMLDB recovery time comparison based on DRAM and persistent memory
Figure 15. OpenMLDB hardware cost comparison based on DRAM and persistent memory

--

--

MemArk

memark.io — Leveraging Modern Storage Architecture for System Enhancement