The problem
In serverless platforms (AWS Lambda, Azure Functions) a function that has not run recently must be loaded first, which adds hundreds of milliseconds to seconds. Keeping functions warm all the time wastes memory and money. The Azure Functions traces show that call patterns are often predictable.
Who needs it
Cloud providers and companies building APIs on serverless; slow first responses hurt user experience.
Idea of the approach
- Use the public Azure Functions traces.
- Train predictors of the next invocation time (histogram policy from the paper as a baseline, then gradient boosting or a small LSTM).
- Simulate keep-alive and pre-warm policies; measure cold-start rate vs wasted memory.
- Optional: test on OpenFaaS or Knative.
What you will build and deliver
- Prediction models and the simulator.
- Trade-off curves.
- Paper.
Skills needed
Python, pandas, machine learning basics.
Research contribution & where to publish
Contribution: learned pre-warming with an explicit memory-cost trade-off. Venues: ACM SoCC, IEEE CLOUD, Middleware, Future Generation Computer Systems.
Key references — start reading here
- M. Shahrad, R. Fonseca, Í. Goiri, G. Chaudhry, P. Batum, J. Cooke, E. Laureano, C. Tresness, M. Russinovich, R. Bianchini. Serverless in the Wild: Characterizing and Optimizing the Serverless Workload at a Large Cloud Provider. USENIX ATC, 2020.
- E. Oakes, L. Yang, D. Zhou, K. Houck, T. Harter, A. Arpaci-Dusseau, R. Arpaci-Dusseau. SOCK: Rapid Task Provisioning with Serverless-Optimized Containers. USENIX ATC, 2018.
- E. Jonas, J. Schleier-Smith, V. Sreekanti, C.-C. Tsai, A. Khandelwal, Q. Pu, et al.. Cloud Programming Simplified: A Berkeley View on Serverless Computing. arXiv, 2019.
Interested?
Send the topic title and the courses you have taken.
Contact Dr. Hamza DjigalOther topics in Edge and cloud scheduling
Budget- and deadline-aware autoscaling of Kubernetes microservices with reinforcement learning
Carbon-aware scheduling of data-pipeline workflows across cloud regions
Spot-instance-aware workflow scheduling to cut cloud costs for data pipelines
A digital twin of a multi-access edge computing network for testing task-offloading policies (open simulator and benchmark)