Loading...
Please wait, while we are loading the content...
Similar Documents
RSI : A High-Efficient and Fault-Tolerant Interconnection for Resources-Pooling in Rack Scale
| Content Provider | Semantic Scholar |
|---|---|
| Author | Lai, Mingche Zou, Xiangxi Jian, Jie Qi, Xingyun Xu, Jiaqing |
| Copyright Year | 2018 |
| Abstract | Data centers are originally composed of servers that deal with different services, and have been developed into rack-mounted computers and blade server systems. At the same time, the traffic between different servers is increasing, and the interconnection network performance between servers has significant influence on the overall performance of the system. Based on the resource pools background in the development of data center recently, this paper proposes a rack-scale interconnection network structure named RSI. According to the features of the interconnection structure, the routing tables are designed with two levels, which can reduce the size of tables and be convenient to realize adaptive routing. Then a low cost adaptive routing called LAR with a threshold to decide the choice of routing is proposed. We come up with a deadlock prevention mechanism and the deadlock can be prevented in LAR with only 2 VCs. In addition, a fault tolerance algorithm is used to deal with potential failures. Compared to current interconnection topologies in a rack, RSI hierarchical topology can support larger scale of nodes with comparable performance. Finally, the evaluation results show that in extreme traffics, LAR achieves about 6 times throughput than minimal routing which performs well in uniform random traffic. Keywords—Rack scale; resource pooling; interconnection; adaptive routing; fault-tolerant |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://saiconference.com/Downloads/FTC2017/Proceedings/155_Paper_381-RSI_A_High-Efficient_and_Fault-Tolerant_Interconnection.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |