From Reliable Models to Resilient ML Platforms
summary
Explains concrete platform choices and operational controls: uses IBM Cloud SoftLayer bare‑metal and virtual compute with a global backbone, recommends multi‑zonal deployments and immutable backups for disaster recovery, and sets SLOs/SLAs for uptime and latency. Details security controls: zero trust, DDoS/ransomware protection, network segmentation, IAM and audit logs, and automated container image scanning as a gate. Recommends rate limits, health checks and automated failover to sustain unpredictable spikes and production drift.
AI-written summary · about
chapters (14)
- 0:00Welcome & Speaker Introduction (Riva at Con 42 20 26)
- 0:24Talk Overview: Moving ML from Lab to Production + Agenda
- 2:59Why Production ML Is Hard: Drift, Scale, Latency & Availability
- 4:02Modern Platforms vs Legacy: Cloud-Native Capabilities
- 4:44IBM Cloud/SoftLayer as an Example Infrastructure Foundation
- 5:33Pillars of Resilient ML Infrastructure: HA & Disaster Recovery
- 7:14Security by Design: Zero Trust, DDoS/Ransomware Protection
- 9:49Sustaining ML Workloads: Rate Limits, Traffic Spikes & DDoS Readiness
- 11:09Segmentation, Environment Isolation & Secure Model Serving
- 13:38Framework Alignment & Operational Controls: IAM, Audit Logs, Image Scanning
- 16:49Performance Metrics & Resiliency Benchmarking (SLOs/SLAs)
- 18:14People & Process: Cross-Functional Ownership for Production ML
- 21:18Deployment Patterns: Cloud-Native vs Hybrid vs Multi-Cloud
- 22:23Design Principles & Key Takeaways + Closing/Q&A
related talks
19mBuilding Resilient ML Cloud Integrations
33mBuilding Resilient (ML) Pipelines for MLOps
29mDemystifying Machine Learning in Production: Reasoning about a Large-Scale ML Platform
17mBuilding Resilient AI Pricing Platforms
30mML Platform for Insurance Conglomerate
58mResilient by Design
18mArchitecting Intelligent Platforms
34mToward continuous resilience