Explore Data Placement Algorithm for Balanced Recovery Load Distribution
summary
Defines optimal recovery-load placement as NP-hard and gives a concrete, implementable greedy data placement algorithm that picks nodes with minimal current edge weight and minimal sum of repair costs to form placement groups. Claims balanced parallel recovery without using fine-grained units, supports low-overhead expansion, and shows 1.7–2.4x faster recovery in benchmarks versus random placement. Evaluations include 500 placement-group scenarios, single- and multi-disk additions, and measured transient imbalances during expansion.
AI-written summary · about
chapters (8)
- 0:00Data placement research intro
- 1:47Recovery load distribution
- 3:59Optimal placement problem
- 5:25NP-hard recovery distribution
- 6:51Greedy balanced recovery
- 7:56System expansion impact
- 9:23Performance gains
- 9:45Summary and future
related talks
13mThe what, The from, and The to: The Migration Games in Deduplicated Systems
31mTradeoffs in Resiliency: Managing the Burden of Data Recoverability
50mParallel Recovery in PostgreSQL
17mParaRC: Embracing Sub-Packetization for Repair Parallelization in MSR-Coded Storage
20mHow Can We Make Data Integrity Easy?
45mData Placement in Linux: Evolving Block and File I/O
17mPractical Design Considerations for Wide Locally Recoverable Codes (LRCs)
47mDisaster Recovery Strategies for SQL Server