Optimizing Spot Instance Savings for Deadline-Sensitive Jobs
Offered By: USENIX via YouTube
Course Description
Overview
Explore a groundbreaking approach to optimizing cloud computing costs while meeting strict deadlines in this award-winning conference talk from NSDI '24. Delve into the challenges of using spot instances for deadline-sensitive jobs and discover a novel solution that combines spot and on-demand instances strategically. Learn about the in-depth characterization of spot instance behavior, including availability, pricing, and duration. Examine the theoretical model developed to analyze worst and average-case scenarios for baseline policies. Gain insights into the Uniform Progress policy, a simple yet effective parameter-free approach that outperforms greedy strategies and narrows the gap to optimal solutions. Understand how this innovative method achieves significant cost savings across various real-world workloads and deadlines when implemented on the SkyPilot intercloud broker system. Access open-sourced spot availability traces to further your research in cloud resource optimization.
Syllabus
NSDI '24 - Can't Be Late: Optimizing Spot Instance Savings under Deadlines
Taught by
USENIX
Related Courses
Communicating Data Science ResultsUniversity of Washington via Coursera Cloud Computing Applications, Part 2: Big Data and Applications in the Cloud
University of Illinois at Urbana-Champaign via Coursera Cloud Computing Infrastructure
University System of Maryland via edX Google Cloud Platform for AWS Professionals
Google via Coursera Introduction to Apache Spark and AWS
University of London International Programmes via Coursera