Unlocking Cross-Organizational Insights: Practical MPC for Cloud-Based Data Analytics
summary
Built a cloud-native MPC pipeline using Carbyne Stack and MP-SPDZ, adding support for a semi-honest model and bits/dits primitives, plus a sorting+bucketing group-by that pads fixed-size arrays. Deployment runs computational parties in one availability zone with zone affinity and spot instances, cutting latency from ~25 ms to ~1 ms and network cost to $0/GB, adds an orchestrator for spot preemption, and reaches a stable, low-cost production deployment for cross-organizational analytics.
AI-written summary · about
chapters (13)
- 0:00Cross-organizational analytics with MPC
- 1:10Adtech use case and privacy
- 2:33Secret shares in MPC
- 3:57Democratizing private analytics
- 5:01MP-SPDZ framework
- 6:24Carbine Stack for cloud MPC
- 8:33Real-world usability enhancements
- 9:36Semi-honest security model
- 10:40VPC peering optimizations
- 11:48Spot nodes for cost savings
- 12:32Group-by for developers
- 13:59Sorting and accumulation algorithms
- 14:20Conclusion and future
related talks
18mSECRECY: Secure collaborative analytics in untrusted clouds
01h 19mWhen Data Can’t Move but Insight Must
47mCloud Native and Sustainable, Reproducible Scientific Computing
18mKeynote: Mind the Gap: Bridging Cloud Native Innovation with Real-World Use
29mProactive and Polymorphic Adaptation of Multi-cloud Deployments
40mMulti Cloud Magic: Leveraging Multi Cloud Clusters in Real World Scenarios
40mApplied MPC
27mFrom Supercomputing to Serving: A Case Study Delivering Cloud Native Foundation Models