Loading...
Please wait, while we are loading the content...
Similar Documents
Efficient Execution of MapReduce Applications on Irregular NoC Topology
| Content Provider | Semantic Scholar |
|---|---|
| Author | Addisie, Abraham Cowan, Meghan Furia, Milind Hagos, Helen |
| Copyright Year | 2015 |
| Abstract | With the exponential growth of data, data-intensive applications are becoming common in CMPs. Such applications are mostly written using MapReduce programming model, because it provides simple programming interface. The underlying interconnect in CMPs is network-on-chip(NoC). Due to extreme device scaling, link failure in NoCs is becoming a common scenario. Link failures leads to an irregular network topology. MapReduce frameworks like Phoenix++ are written based on an assumption of a robust regular network topology. For communication bound applications, irregular network topology leads to inefficient execution of MapReduce applications. Both the mapping phase and reducing phase of MapReduce applications provide opportunity for load balancing based on the connectivity of a node to the rest of the system. In this project, we proposed and evaluated novel load balancing algorithm in both phases of MapReduce. We found that depending on the communication to computation ratio and the number of unique keys of workloads, the execution time and network latency can be significantly improved. |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://www.eecs.umich.edu/courses/eecs578/eecs578.f15/projects/HAMM_outline.pdf |
| Alternate Webpage(s) | http://www.eecs.umich.edu/courses/eecs578/eecs578.f15/projects/HAMM_checkpoint3.pdf |
| Alternate Webpage(s) | http://www.eecs.umich.edu/courses/eecs578/eecs578.f15/projects/HAMM_poster.pdf |
| Alternate Webpage(s) | http://www.eecs.umich.edu/courses/eecs578/eecs578.f15/projects/HAMM_report.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
