Adaptive Packet Throttling Technique for Congestion Management in Mesh NoCs
| Autor: | John Jose, R. S. Reshma Raj, N. S. Aswathy, Abhijit Das, V. R. Josna |
|---|---|
| Rok vydání: | 2017 |
| Předmět: |
010302 applied physics
Computer science Network packet business.industry Packet injection Spec# 02 engineering and technology Congestion management Bandwidth throttling 01 natural sciences 020202 computer hardware & architecture Network congestion Network on a chip 0103 physical sciences 0202 electrical engineering electronic engineering information engineering Latency (engineering) business computer Computer network computer.programming_language |
| Zdroj: | Communications in Computer and Information Science ISBN: 9789811074691 VDAT |
| DOI: | 10.1007/978-981-10-7470-7_33 |
| Popis: | Network on Chip is an emerging communication framework for multi-core systems. Due to increasing number of cores and complex workloads, congestion management techniques in NoC are gaining more research focus. Packet throttling is one of a cost effective technique for congestion management. It delays the packet injection into the network, thereby regulating traffic in network and hence provide ease of packet movement generated by other critical applications. Finding point of throttling and rate of throttling are two major design issues that can impact the performance and stability of any throttling algorithm. Existing state of the art throttling techniques use local throttling decision coordinated by a single central controller. We overcome the issues related with this by partitioning the network into number of subnetworks, each with a zonal controller. Our experiment results in 8 \(\times \) 8 2D mesh with real traffic workloads consisting of SPEC 2006 CPU benchmarks shows an average packet latency reduction of 6.2% than the state of the art packet throttling techniques. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|