.png){kind=icon}
.png){kind=icon}
Return
GMR: graph-compatible MapReduce programming model
.png){kind=icon}
PRE
AI
DOI：10.1007/s11042-017-5102-2.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The MapReduce programming model is widely used to parallelize data processing over the large scale of commodity computer clusters. However, on account of its monotonous data representation, it fails to express graph-parallel algorithms naturally and execute them efficiently. Alternatively, Pregel and PowerGraph could address these challenges. But they require users to familiarize another set of programming patterns and platforms, and at the same time the legacy MapReduce code also becomes incompatible and useless. In this paper, we proposed the Graph-compatible MapReduce (GMR) as an extension of Google's Standard MapReduce (SMR). In this way, graph-parallel algorithm will be naturally expressed without compromising the efficiency and simplicity, and meanwhile the conventional MapReduce programming pattern be preserved. Also, users could gain the convenience of Think like a vertex. Based on the experimental studying, we analyzed the ratio of the redundant computation, transmission and data caching introduced in naive iterative MapReduce platforms (e.g., HaLoop, Twister). Furthermore, we discussed the difference between GMR and the graph-targeted frameworks. The evaluation experiment results show that GMR outperforms GraphX in a series of real-world graph-parallel algorithms.
Keywords:
Distributed systems
Parallel architectures
Graph theory
Systems programs and utilities
Performance analysis and design aids
Concurrent programming
Modes of computation
Performance of systems
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
3
Papers:
1.9W
Citations:
3.2W