SHF: Small: Collaborative Research:

Elastic Fidelity: Trading-Off Computational Accuracy for Energy Efficiency

Funding Agency: National Science Foundation

Directorate: Computer & Information Science & Engineering (CISE)

Division: Computer and Communication Foundations (CCF)

Program: Software and Hardware Foundations (SHF)

Award Numbers: 1218768 (Northwestern University) and 1217353 (The Ohio State University)

Program Manager: Hong Jiang (2013-2015), Ahmed Louri (2012)

PIs: Nikos Hardavellas (Northwestern University), Seda Ogrenci Memik (Northwestern University), Srinivasan Parthasarathy (The Ohio State University)

Institutions: Department of Electrical Engineering and Computer Science, Northwestern University, and Department of Computer Science and Engineering, The Ohio State University

Affiliated Graduate Students: Georgios Tziantzioulis (Northwestern University), Ali Murat Gok (Northwestern University), Yigit Demir (Northwestern University), and S.M. Faisal (The Ohio State University)

Recently Graduated Students: Yigit Demir (Ph.D., Northwestern), S.M. Faisal (Ph.D. The Ohio State University), Xinxin Huang (M.S., Northwestern), Ke Liu (M.S., Northwestern), Sourya Roy (Honors Undergraduate, Northwestern).

Project Dates: August 1, 2012 to July 31, 2015

Abstract:

Energy and power consumption have become a critical issue ranging from microarchitectures to large-scale data centers and supercomputers. Conservative estimates suggest that the information technology industry world-wide energy consumption is in excess of 400 TWh and growing, generating roughly the same carbon footprint as the airline industry, accounting for 2% of global emissions. At the same time, the power constraints of chips hamper their performance, and the shrinking transistor geometries and low supply voltages increase the severity of processor variations resulting in higher timing error rates. High error rates lead to a significant drop in yield and increased manufacturing costs, calling for designs that are able to withstand them.

This project seeks to understand and explore the novel paradigm of elastic fidelity computing. Elastic fidelity computing capitalizes on the observation that many applications can naturally tolerate errors, and that not all of them need to run at 100% fidelity all the time. Specifically, the goal of this work is to understand the error models of various hardware components as they relate to data movement, storage, and computation, and simultaneously to understand the error resiliency of applications and re-architect them to leverage elastic fidelity.

Elastic fidelity offers potentially transformative effects for science and society, by challenging conventional wisdom and taking a fresh look at the interplay of errors, output quality and energy efficiency for an important class of pervasive streaming and data-intensive applications. More specifically, elastic fidelity promises significant energy savings that can put computing on an environmentally sustainable path, by lowering the operational costs in major economic sectors, and making the manufacturing of future chips cheaper by relaxing the accuracy requirements of hardware components. The results of this research will be disseminated through publications, workshops, advanced curriculum, and releases of the developed infrastructure in the public domain. To accelerate broad societal effects, the project participants will seek to foster technology transfer by promoting collaboration and industry involvement through presentations and site visits.

Project Page: http://www.eecs.northwestern.edu/~hardav/projects/elastic_fidelity/

Dataset/Software Release:

1. SoftInj: a software fault injection library that implements the b-HiVE error models.

2. b-HiVE Hardware Characterization: raw experimental data from HSIM and SPICE simulations of 64-bit integer ALUs, integer multipliers, bitwise logic operations, FP adders, FP multipliers, and FP dividers from OpenSparc T1, along with controlled value correlation experiments.

3. Gem5 patches for Elastic Fidelity: implementation of b-HiVE error models and Lazy Pipelines within Gem5.

4. LLVM patches for Elastic Fidelity: extensions to the LLVM infrastructure to support the Elastic Fidelity ISA extensions and compile applications for execution in our version of Gem5.

Publications:

1. Energy Proportional Photonic Interconnects
Y. Demir and N. Hardavellas. Preprint, 2016 (in preparation)

2. Divergent Flattened Butterfly: A novel photonic layout for Flattened Butterfly Networks
Y. Demir and N. Hardavellas. Preprint, 2016 (in preparation)

3. IMF: Thermally Stable Nanophotonic Interconnects through Insulation and Micro-Fluidics
Y. Demir, N. Hardavellas and D. Atienza. Preprint, 2016 (in preparation)

4. Laser Power Gating for Datacenter Interconnects
Y. Demir, M. Sanchez, H. Han, S. Kandula, F. Bustamante, and N. Hardavellas. Preprint, 2016 (in preparation)

5. SLaC: Stage Laser Control for a Flattened Butterfly Network
Y. Demir and N. Hardavellas. In Proceedings of the 22nd IEEE Symposium on High Performance Computer Architecture (HPCA), Barcelona, Spain, March 2016

6. Lazy Pipelines: Enhancing Quality in Approximate Computing
G. Tziantzioulis, A. M. Gok, S. M. Faisal, N. Hardavellas, S. Memik, and S. Parthasarathy. In Proceedings of the Design, Automation, and Test in Europe (DATE), Dresden, Germany, March 2016

7. Edge Importance Identification for Energy Efficient Graph Processing
S. Faisal, G. Tziantzioulis, A. Gok, S. Parthasarathy, N. Hardavellas, S. Ogrenci-Memik. In IEEE BigData, Santa Clara, CA, October 2015

8. A Methodology for Power Characterization of Associative Memories
Dawei Li, Siddhartha Joshi, Seda Ogrenci-Memik, James Hoff, Sergo Jindariani, Tiehui Liu, Jamieson Olsen and Nhan Tran. In Proceedings of the 33rd IEEE International Conference on Computer Design (ICCD), New York City, NY, October 2015

9. SCP: Synergistic Cache Compression and Prefetching
B. Patel, G. Memik and N. Hardavellas. In Proceedings of the 33rd IEEE International Conference on Computer Design (ICCD), New York City, NY, October 2015

10. b-HiVE: A Bit-Level History-Based Error Model with Value Correlation for Voltage-Scaled Integer and Floating Point Units
G. Tziantzioulis, A. M. Gok, S. M. Faisal, N. Hardavellas, S. Memik, and S. Parthasarathy. In Proceedings of the Design Automation Conference (DAC), San Francisco, CA, June 2015

11. Towards Energy-Efficient Photonic Interconnects
Y. Demir and N. Hardavellas. In Proceedings of SPIE, Optical Interconnects XV, San Francisco, CA, February 2015. Also selected to appear in SPIE Green Photonics, 2015

12. Automatic Selection of Sparse Matrix Representation on GPUs
Naser Sedaghati, Te Mu, Louis-Noel Pouchet, Srinivasan Parthasarathy, P. Sadayappan. ICS 2015: 99-108

13. Sequential Hypothesis Tests for Adaptive Locality Sensitive Hashing
Aniket Chakrabarti, Srinivasan Parthasarathy. WWW 2015: 162-172

14. One, Two, Hash! Counting Hash tables for SSD devices
Tyler Clemons, S. M. Faisal, Shirish Tatikonda, Charu C. Aggarwal, and Srinivasan Parthasarathy. IKDD Conference on Data Sciences (CoDS), New Delhi, India, March 2014

15. Fast Sparse Matrix-Vector Multiplication on GPUs for Graph Applications
Arash Ashari, Naser Sedaghati, John Eisenlohr, Srinivasan Parthasarathy, and P. Sadayappan. In Proceedings of Supercomputing'14: International Conference for High Performance Computing, Networking, Storage and Analysis, New Orleans, LA, November 2014

16. A Bayesian Perspective on Locality Sensitive Hashing with Extensions for Kernel Methods
A. Chakrabarti, V. Satuluri, S. Parthasarathy. VLDB Journal, 2014.

17. A fast implementation of MLR-MCL algorithm on multi-core processors
Qingpeng Niu, Pai-Wei Lai, S. M. Faisal, Srinivasan Parthasarathy, P. Sadayappan. HiPC 2014: 1-10

18. LaC: Integrating Laser Control in a Photonic Interconnect
Y. Demir and N. Hardavellas. In Proceedings of the IEEE Photonics Conference (IPC), La Jolla, CA, October 2014

19. EcoLaser: An Adaptive Laser Control for Energy-Efficient On-Chip Photonic Interconnects
Y. Demir and N. Hardavellas. In Proceedings of the International Symposium on Low Power Electronics and Design (ISLPED), La Jolla, CA, August 2014

20. Galaxy: A High-Performance Energy-Efficient Multi-Chip Architecture Using Photonic Interconnects
Y. Demir, Y. Pan, S. Song, N. Hardavellas, G. Memik and J. Kim. In Proceedings of the ACM International Conference on Supercomputing (ICS), Munich, Germany, June 2014

21. LaC: Integrating Laser Control in a Photonic Interconnect
Y. Demir and N. Hardavellas. Technical Report NU-EECS-14-03, Northwestern University, Evanston, IL, April 2014

22. EcoLaser: An Adaptive Laser Control for Energy Efficient On-Chip Photonic Interconnects
Y. Demir and N. Hardavellas. Technical Report NU-EECS-14-02, Northwestern University, Evanston, IL, April 2014

23. Hash in a flash: Hash tables for flash devices
Tyler Clemons, S. M. Faisal, Shirish Tatikonda, Charu C. Aggarwal, and Srinivasan Parthasarathy. In Proceedings of the IEEE International Conference on BigData, pp. 7-14, Santa Clara, CA, October 2013

24. The Impact of Dynamic Directories on Multicore Interconnects
M. Schuchhardt, A. Das, N. Hardavellas, G. Memik, and A. Choudhary. IEEE Computer, Special Issue on Multicore Memory Coherence, Vol. 46(10), October 2013

25. Galaxy: A High-Performance Energy-Efficient Multi-Chip Architecture Using Photonic Interconnects
Y. Demir, Y. Pan, S. Song, N. Hardavellas, J. Kim, and G. Memik. Technical Report NU-EECS-13-08, Northwestern University, Evanston, IL, July 2013

26. Network Clustering (Chapter 17)
S. Parthasarathy and S. M. Faisal. Data Clustering, Algorithms and Applications. Eds. C. Aggarwal and C. Reddy. ISBN-13: 978-1466558212. Chapman and Hall/CRC Press, August 2013

27. Elastic Fidelity: Trading-off Computational Accuracy for Energy Reduction
Sourya Roy, Tyler Clemons, S. M. Faisal, Ke Liu, Nikos Hardavellas, and Srinivasan Parthasarathy. Technical Report NWU-EECS-11-02, Northwestern University, Evanston, IL, February 2011. Indexed at arXiv:1111.4279 [cs.AR], November 2011 (published prior to grant award)

Ph.D., M.S., and Undergraduate Theses:

1. Harnessing Approximation and Heterogeneity for Energy- and Power-Efficient Computing
Georgios Tziantzioulis. Ph.D. Proposal. Department of Electrical Engineering and Computer Science, Northwestern University, Evanston, IL, USA, October 2015

2. High-Performance and Energy-Efficient Computer System Design Using Photonic Interconnects
Yigit Demir. Ph.D. Thesis. Department of Electrical Engineering and Computer Science, Northwestern University, Evanston, IL, USA, August 2015

3. Towards Energy Efficient Data Mining and Graph Processing
S. Faisal. Ph.D. Thesis. Department of Computer Science and Engineering, The Ohio State University, Columbus, Ohio, 2015

4. Energy Efficient Data Placement and Algorithms
Aniket Chakrabarty. Ph.D. Candidacy Proposal. Department of Computer Science and Engineering, The Ohio State University, Columbus, Ohio, 2015

5. High-Performance and Energy-Efficient Computer System Design Using Photonic Interconnects
Yigit Demir. Ph.D. Proposal. Department of Electrical Engineering and Computer Science, Northwestern University, Evanston, IL, USA, 2014

6. Modeling the Impact of Process and Thermal Variations and Materials on Nanophotonic Devices
Xinxin Huang. M.S. Thesis. Department of Electrical Engineering and Computer Science, Northwestern University, Evanston, IL, USA, May 2013

7. Towards Energy Efficient Data Mining and Graph Processing
S. Faisal. Ph.D. Candidacy Proposal. Department of Computer Science and Engineering, The Ohio State University, Columbus, Ohio, 2015

8. Hardware Error Rate Characterization with Below-Nominal Supply Voltages
Ke Liu. M.S. Thesis. Department of Electrical Engineering and Computer Science, Northwestern University, Evanston, IL, USA, December 2012

9. Elastic Fidelity: Trading-off Computational Accuracy for Energy Reduction
Sourya Roy. Honors Thesis, Northwestern University, Evanston, IL, March 2011 (published prior to grant award)

Disclaimer: This material is based upon work supported by the National Science Foundation under Grant Number CCF-1218768 and CCF-1217353. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.