Reproduce “Prediction-driven resource provisioning for serverless container runtimes” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Dimitrios Tomaras, Michail Tsenos, Vana Kalogeraki
SLOPE, a prediction framework for serverless Function-as-a-Service platforms, proactively provisions resources by combining neural networks for known applications and graph similarity for novel ones. This approach minimizes cold start latencies and meets Service-Level Objectives for dynamic workloads, reducing operational costs by an average of 66.25% compared to naive overprovisioning.
Key insights
- Categorical Cross-Entropy (CCE) consistently yields high accuracy and low loss in neural network-based resource prediction for serverless function replica allocation.
- Serverless applications with similar codebase structures (represented by call graphs) exhibit similar performance characteristics, allowing for resource inference for novel functions via Graph Edit Distance.
- Proactive resource provisioning, as implemented by SLOPE, can achieve substantial operational cost reductions while meeting performance SLOs more effectively than reactive auto-scaling or overprovisioning.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.