On the Impact of Isolation Costs on Locality-aware Cloud Scheduling
summary
Argues that embedding functions into storage saves data movement but pays nontrivial isolation overheads, so a scheduler must profile data movement and network costs, measure three isolation schemes and their latencies, and use a modular cost model to decide placement. Describes a prototype scheduler that instruments functions, quantifies isolation and transfer costs across real serverless workloads, shows isolation can often erase locality benefits, and that model-driven placement recovers most efficiency without changing applications.
AI-written summary · about
chapters (5)
- 0:00Introduction and context
- 3:54Impact of storage services
- 8:53Local data system breakdown
- 12:22Isolation cost schemes
- 15:50Q&A and summary
related talks
15mHere, There and Everywhere: The Past, the Present and the Future of Local Storage in...
21mHarmonizing Efficiency and Practicability: Optimizing Resource Utilization in...
05mSponsored Keynote: The Cloud (And Your $$$) is Not Infinite: Dynamic Scheduling at Every Layer
45mEmbracing Big Data Workload in Cloud-Native Environment with Data Locality
22mUnFaaSener: Latency and Cost Aware Offloading of Functions from Serverless Platforms
26mShuffling, Fast and Slow: Scalable Analytics on Serverless Infrastructure
43mDo modern cloud applications lock you in?
58mServerless: Five Key Things you need to Know