Abstract | ||
---|---|---|
As an emerging technology to empower on-chip communication of future many-core processors, optical crossbars achieve low uniform latency by providing non-blocking one-hop connectivity. Conventional token-based arbitration limits throughput by token injection rates so that large throughput are delivered by over provisioning optical channels, consuming significant optical power. In this paper, we propose pipelined distributed global arbitration for multiple-write multiple-read optical crossbars to improve energy efficiency without downgrading performance. Multiple requests for the same channel can be granted at the same cycle as long as no traffic conflicts occur. The proposed global arbitration scheme enables a scalable arbiter design which parallelizes the arbitration process and simplifies arbiter logic for low arbitration latency, low power and high network throughput. Our simulation results show that, the proposed design reduces execution time by 5% and power consumption by 20% using only 50% of channels and 75% of micro rings as in the best known baseline. |
Year | DOI | Venue |
---|---|---|
2015 | 10.1109/IPDPS.2015.61 | International Parallel & Distributed Processing Symposium |
Keywords | Field | DocType |
optical networks, arbitration | Arbiter,Computer science,Efficient energy use,Latency (engineering),Parallel computing,Computer network,Communication channel,Arbitration,Throughput,Security token,Distributed computing,Scalability | Conference |
ISSN | Citations | PageRank |
1530-2075 | 3 | 0.40 |
References | Authors | |
22 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Jiwei Liu | 1 | 6 | 3.48 |
Jun Yang | 2 | 366 | 22.63 |
Rami G. Melhem | 3 | 240 | 16.11 |