Caerus: NIMBLE Task Scheduling for Serverless Analytics
summary
Caerus is a task-level scheduler that implements the NIMBLE algorithm to schedule serverless analytics tasks by computing per-task launch times. It models substage step model pipelineability, uses an input estimator and runtime profiler, and applies a scheduling module to enforce launch times. NIMBLE provably minimizes cost and is Pareto-optimal between job completion time and cost; the Caerus implementation evaluates on TPC-DS and other big-data queries and reports achieving both near-optimal JCT and minimal cost.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 0:25Serverless
- 0:53Performance Goals
- 1:42Scaling Problem
- 2:40MapReduce Example
- 4:25Main Idea
- 5:14Challenges
- 5:55Pipelineability
- 6:30Overview
- 6:56Stem Model
- 9:00Algorithm
- 9:44Over Analysis
- 10:19Series
- 10:54Conclusion
related talks
21mHarmonizing Efficiency and Practicability: Optimizing Resource Utilization in...
18mALPS: An Adaptive Learning, Priority OS Scheduler for Serverless Functions
22mUnFaaSener: Latency and Cost Aware Offloading of Functions from Serverless Platforms
11mLightning Talk: Searching for Serverless Candidates in the Jungle of Microservices
16mTowards Plan-aware Resource Allocation in Serverless Query Processing
01h 00mServerless - reality or BS - notes from the trenches
45mAutonomous Multi-Cloud Serverless Deployment and Optimized Management
52mImplement a Serverless Data Analytics strategy