FlyOver

FlyOver (FLOV) is a voting approach for adaptive router power-gating in a light-weight and distributed manner, which includes router microarchitecture enhancement, adaptive power-gatiing policy, and low-latency dynamic routing algorithms. More details can be found in our and IEEE TC 2020 paper and IPDPS'17 paper:

Jiayi Huang, Shilpa Bhosekar, Rahul Boyapati, Ningyuan Wang, Byul Hur, Ki Hwan Yum, and Eun Jung Kim, "A Voting Approach for Adaptive Network-on-Chip Power-Gating", in IEEE Transactions on Computers, vol. 70, no. 11, pp. 1962-1975, 2021.

Rahul Boyapati*, Jiayi Huang*, Ningyuan Wang, Kyung Hoon Kim, Ki Hwan Yum, and Eun Jung Kim, "Fly-Over: A Light-Weight Distributed Power-Gating Mechanism for Energy-Efficient Networks-on-Chip", in Proceedings of the IEEE International Parallel and Distributed Processing Symposium (IPDPS), pp. 708-717, 2017.

About This Repository

This repository contains the implementation of FlyOver, a distributed network-on-chip power-gating mechanism. BookSim 2.0 is used as the network-on-chip simulator for both synthetic and full-system simulation. The booksim2 directory contains the code to run synthetic workloads. The gem5 directory has the gem5 with BookSim 2.0 integrated for full-system simulations.

FlyOver Implementation

The main components of FLOV are implemented in the BookSim 2.0 simulator:

• Implementing the voting policy in a new traffic manager class FLOVTrafficManager inherited from TrafficManager to collect votes from all the routers and make the transitions of power-gating policy for individual routers (among No-FLOV, R-FLOV, and G-FLOV).
• Enhancing the Network and KNCube classes to support power-gating configuration and to add handshake channels for power-gating protocols.
• Introducing FLOVRouter router class that inherits from IQRouter to implement the introduce router microarchitecture changes for flyover bypassing, credit relaying of flow control and power-gating handshaking.
• Implementing two dynamic deadlock-free FLOV routing algorithms flov_mesh and adaptive_flov_mesh in routefunc.cpp to faciliate low-latency packet delivery using FlyOver links for bypassing in irregular power-gated networks.
• Adding the DSENT_POWER_Module class for parameterized power model, whose energy and power parameters are obtained from DSENT power model (see utils/dsent_models.cfg).

Running Simulations

You can check booksim2/README.md and gem5/README.md for instructions to run synthetic and full-system simulations.

---

If you find this useful in your research, please consider citing our work:

@article{huang:tc:2020:flyover,
 author = {Huang, Jiayi and Bhosekar, Shilpa and Boyapati, Rahul and Wang, Ningyuan and Hur, Byul and Yum, Ki Hwan and Kim, Eun Jung},
 title = {{A Voting Approach for Adaptive Network-on-Chip Power-Gating}},
 jornal = {IEEE Transactions on Computers},
 year = {2021},
 volume = {70},
 number = {11},
 pages = {1962--1975},
 doi = {10.1109/TC.2020.3033163}
}

@inproceedings{boyapati:ipdps:2017:flyover,
 author = {Boyapati, Rahul and Huang, Jiayi and Wang, Ningyuan and Kim, Kyung Hoon and Yum, Ki Hwan and Kim, Eun Jung},
 title = {{Fly-Over: A Light-Weight Distributed Power-Gating Mechanism For Energy-Efficient Networks-on-Chip}},
 booktitle = {Proceedings of the 2017 International Parallel and Distributed Processing Symposium (IPDPS)},
 pages = {708--717},
 month = {May},
 year = {2017}
}