HiDPU: A DPU-Oriented Hybrid Indexing Scheme for Disaggregated Storage Systems

on YouTube ↗18m

" summary

HiDPU is a DPU-oriented hybrid index that partitions mapping entries into accurate, PTHash, and LPTHash segments, builds a layered learned index across segment first-keys, and keeps small upper-layer indexes and a pilot cache on the DPU to cut host interactions. A two-phase asynchronous update stalls inserts in a CPU buffer and performs local segment rebuilds; a prototype on a Huawei Hi1823 DPU (4 MB internal memory) reports up to 92% memory savings and up to 6.3x query improvement.

AI-written summary · about

" chapters (10)

1. 0:00HiDPU and DPU storage intro
2. 2:06CPU-centric address translation challenges
3. 3:52DPU storage mapping limits
4. 4:35HiDPU hybrid indexing design
5. 8:18Index updates and CPU-DPU collaboration
6. 10:11HiDPU prototype implementation
7. 13:06Performance evaluation vs baselines
8. 14:41HiDPU scalability and memory overhead
9. 15:46Index reconstruction impact
10. 17:13Conclusion and future work