UnFaaSener: Latency and Cost Aware Offloading of Functions from Serverless Platforms
summary
Framework harvests spare VMs and on‑prem hosts and routes individual functions off the serverless path using pub/sub glue and a tiny injected header that reads routing decisions, enabling partial per-function offloading. An asynchronous scheduler predicts host availability and latency, feeds a mixed integer non-linear solver that balances cost versus added latency. Evaluation on image and video workflows shows up to ~89.8% cost reduction under a 90% offload cap, with latency mode reducing that savings but keeping added delay near 10% in some workloads.
AI-written summary · about
chapters (5)
- 0:00The rise of serverless
- 1:55Proposal for function offloading
- 6:46System architecture design
- 11:35Optimization and scheduling logic
- 18:01Evaluation and conclusion
related talks
53mServerless is the Abstraction We Deserve
23mFaasm: Lightweight Isolation for Efficient Stateful Serverless Computing
23mOpenFaaS: Serverless platform with no strings attached?
37mGo FaaSter: Cold Start Optimization in a Serverless Platform
19mDiscovering serverless with OpenFaaS
51mFaaS or not to FaaS? Visible and invisible benefits of the Serverless paradigma
42mFaaS or not to FaaS: Visible & invisible benefits of the Serverless paradigm
33mGoing FaaSter: Cost-Performance Optimizations of Serverless on Kubernetes