GraphWalker: An I/O-Efficient and Resource-Friendly Graph Analytic System...
summary
System design claims an state-aware I/O model plus asynchronous walk updating and block-centric indexing that let a single 64 GB machine with RAID‑0 SSDs process disk‑resident graphs (hundreds of billions of edges) and tens of millions of long walks. Prototype reduces IOs (115 → 49 in a 1.6B‑vertex Yahoo graph), finishes 10M walks in ~1 hour, and reports 17–60× speedups over DrunkardMob and up to 2–40× vs Graphene/GraFSoft while matching distributed KnightKing performance on many workloads.
AI-written summary · about
chapters (16)
- 0:00Intro
- 0:23Graph analytics
- 0:47Large graph scale
- 2:26Graph processing framework
- 3:02Random walk efficiency
- 3:30#walkers in loaded subgraphs
- 6:04#steps walkers can update
- 9:17Walks storage management - walk indexing
- 10:17Walks storage management - walk storing
- 11:57Other optimizations and prototype system
- 12:41Evaluation settings
- 13:43Comparison systems
- 14:35Performance study in entire design space
- 17:25Micro-benchmarks
- 17:56Comparison with state-of-the-art systems
- 19:10Conclusion
related talks
18mSOWalker: An I/O-Optimized Out-of-CoreGraph Processing System for Second-Order...
32mLarge-Scale Graph Processing on Emerging Storage Devices
50mBig Data on Smallish Machines
09mEfficient Large Graph Processing with Chunk-Based Graph Representation Model
27mData @scale Seattle - Yucheng Low
39mThorny path to the Large Scale Graph Processing
27mGraphOne: A Data Store for Real-time Analytics on Evolving Graphs
01h 00mGraph Mastery to AI