Adaptive Placement for In-memory Storage Functions
summary
A system implements a server-internal cost model that observes user-defined functions and dynamically chooses client or storage execution for each invocation on an in-memory key-value store. The prototype runs strict serializable storage functions and achieves 3.2M UDFs/s with ~100 μs responses, improves mixed-workload throughput 85% over server-only and 10% over client-only execution, and cuts latency up to 2x and transactional aborts up to 33%.
AI-written summary · about
chapters (6)
- 0:00Introduction to talk
- 3:24Storage functions challenges
- 8:29Client vs server placement
- 13:15Policy mechanisms for placement
- 20:15Performance experiments results
- 24:26Migration in mobile systems
related talks
45mBeyond Machine Boundaries: Experience with Memory Disaggregation
16mFUSEE: A Fully Memory-Disaggregated Key-Value Store
16mA Tale of Two Paths: Toward a Hybrid Data Plane for Efficient Far-Memory Applications
19mBuilding a Performant Configuration-Store Layer
35mSystems @Scale Summer Q&A Week 3
17mDuhu: Shared Disaggregated Memory for Distributed Data Processing Frameworks
23mSize-aware Sharding For Improving Tail Latencies in In-memory Key-value Stores
29mUnification of Temporary Storage in the NodeKernel Architecture