Panel: Databases and Hardware
summary
Panel reports a measured ~1% utilization on sparse arrays when running real sparse kernels, describes an Onyx tape-out and a working Sam prototype, and shows lossless compression pushing beyond peak bandwidth while research now counts ~350 encoding schemes. Compiled and vectorized engines reach similar speed but compilation adds complexity and code‑explosion risk; cardinality estimation remains unsolved for sparse joins despite histogram work, and cloud abstractions (S3 layer) shift the bottleneck to cross‑node communication.
AI-written summary · about
chapters (11)
- 0:00Panel introduction
- 1:50Cross-disciplinary insights
- 3:37Hardware lessons for PL and DB
- 5:00DB hardware customization
- 7:28Bandwidth compression and bottlenecks
- 10:39Cloud hardware abstraction
- 11:43DB execution compilation trends
- 13:50Industrial code generation challenges
- 15:38Sparsity and cardinality challenges
- 16:41Sparse tensor scheduling problems
- 17:22Closing remarks
related talks
28mBeyond Parquet and ORC: Upgrading Data Infrastructure for Multi-modal AI with Lance Col...
18mThe "New Normal" or An Evolution
26mLeveraging Moore's Law to Optimize Database Performance
46mKeynote: When Hardware and Databases Collide
26mLeveraging Moore's Law to Optimize Database Performance
27mPreview: Storage
48mAnalytics Architectures on Microsoft Fabric
14mRe-think Data Management Software Design Upon the Arrival of Storage Hardware with