Very Large Scale Graph Processing
Event details
| Date | 08.06.2016 |
| Hour | 14:30 › 16:30 |
| Speaker | Laurent Bindschaedler |
| Location | |
| Category | Conferences - Seminars |
EDIC Candidacy Exam:
Exam President: Prof. Edouard Bugnion
Thesis Director: Prof. Willy Zwaenepoel
Co-examiner: Prof. Christos Kozyrakis
Background papers:
Pregel: A System for large-scale Graph Processing
Powergraph: Distributed Graph-parallel Computation on
Natural Graphs
Graphchi: Large-scale Graph Computation on just a PC.
Abstract
Analytics over large graphs is an important and challenging problem due to the large scale and structure of many graphs. Current approaches to tackle this problem either scale out to multiple machines or scale up on a single machine using external memory. We survey existing systems exemplifying these trends. Pregel introduces a general-purpose abstraction based on local vertex computation. PowerGraph refines the computation model to reduce overhead and load imbalance. GraphChi rearranges graph structure to process them efficiently out-of-core. Finally, we investigate a third design direction: scaling out graph computation using secondary storage. We present Chaos and demonstrate its scalability by processing a graph input of 250 terabytes on 20 commodity servers and provide directions for future research.
Exam President: Prof. Edouard Bugnion
Thesis Director: Prof. Willy Zwaenepoel
Co-examiner: Prof. Christos Kozyrakis
Background papers:
Pregel: A System for large-scale Graph Processing
Powergraph: Distributed Graph-parallel Computation on
Natural Graphs
Graphchi: Large-scale Graph Computation on just a PC.
Abstract
Analytics over large graphs is an important and challenging problem due to the large scale and structure of many graphs. Current approaches to tackle this problem either scale out to multiple machines or scale up on a single machine using external memory. We survey existing systems exemplifying these trends. Pregel introduces a general-purpose abstraction based on local vertex computation. PowerGraph refines the computation model to reduce overhead and load imbalance. GraphChi rearranges graph structure to process them efficiently out-of-core. Finally, we investigate a third design direction: scaling out graph computation using secondary storage. We present Chaos and demonstrate its scalability by processing a graph input of 250 terabytes on 20 commodity servers and provide directions for future research.
Practical information
- General public
- Free
Contact
- Cecilia Chapuis EDIC