Crack the FaaS Cold Start and Scalability Bottleneck
summary
Prototype implements snapshot-based instantiation for Firecracker/Kata microVMs and reduces measured cold-start from about 1700 ms to 630 ms (roughly 60% cold-start reduction). Snapshots are taken immediately after the function manager starts to avoid secrets and unique-instance IDs; a small regeneration step reassigns instance-specific data at startup. Function images split into essential code blocks and non-essential parts so download + snapshot (about 50+ MB snapshots from 128 MB VMs) is much smaller and starts execution earlier. Autoscaling speeds up by resizing existing microVM resource limits and spawning additional containers inside running sandboxes.
AI-written summary · about
chapters (5)
- 0:00Start
- 0:10Crack the FaaS Cold Start and Scalability Bottleneck
- 9:33Challenges of Creating A New Function Instance Fromis Snapshot
- 16:36Testbed
- 19:42Example tunction startup breakdown
related talks
27mServerless MicroVMs for Containerized Applications
20mFirecracker: Lightweight Virtualization for Serverless Applications
33mLanguage Runtimes for the New Cloud Era
24mUsing Unikernels to deploy real-world apps
15mSabre: Hardware-Accelerated Snapshot Compression for Serverless MicroVMs
30mLambda SnapStart – CRaC’ing the cold start
24mLightweight VMs for serverless and containers
20mExpeditious High-Concurrency MicroVM SnapStart in Persistent Memory with an...
