The increasing adoption of cloud computing and the rise in popularity of resource-intensive Big Data and Machine Learning workloads has led to a steady growth in data center deployments worldwide, raising fears of excessive energy consumption and greenhouse gas emissions. Given that the benefits of existing hardware and virtualization technologies have largely plateaued, new, disruptive approaches are needed to address this "second wave" of data center energy concerns.
This project proposes data center management techniques that will enable "sustainable-by-design" data centers. The key component of our work will be an end-to-end secure sustainability accounting framework that estimates the total sustainability footprint of a job across all layers of the data center stack that the job traverses. Our framework will take into account the cleanliness of the energy source, its greenhouse gas emissions, and the embedded energy of IT equipment, thus holistically capturing sustainability costs. To realize end-to-end benefits, we will develop new programming models and tools that allow users to participate in sustainability efforts by securely exposing job-specific information and control knobs to data center operators. At runtime, the job sustainability cost estimates and user-provided information will guide data center management components, such as the scheduler and resource allocator, to make data-center–wide sustainable choices.
Investigators at Stony Brook:
Collaborators:
Publications:
- Empirical Evaluation of ML Models for Per-Job Power Prediction
Debajyoti Halder, Manas Acharya, Aniket Malsane, Anshul Gandhi, Erez Zadok
HotCloudPerf 2024 [pdf]
- Verifiable Sustainability in Data Centers
Syed Rafiul Hussain, Patrick McDaniel, Anshul Gandhi, Kanad Ghose, Kartik Gopalan, Dongyoon Lee, Yu David Liu, Zhenhua Liu, Shuai Mu, Erez Zadok
IEEE Security & Privacy Magazine [pdf]
- OpperTune: Post-Deployment Configuration Tuning of Services Made Easy
Gagan Somashekar, Karan Tandon, Anush Kini, Ranjita Bhagwan, Anshul Gandhi, Mayukh Das, Nagarajan Natarajan, Petr Husak, Chieh-Chun Chang
NSDI 2024 [pdf]
- Metrics for Sustainability in Data Centers
Anshul Gandhi, Kanad Ghose, Kartik Gopalan, Syed Rafiul Hussain, Dongyoon Lee, Yu David Liu, Zhenhua Liu, Patrick McDaniel, Shuai Mu, Erez Zadok
ACM SIGEnergy Energy Informatics Review [pdf]
- Evaluating the energy impact of device parameters for DNN inference on edge
Anurag Dutt, Sri Pramodh Rachuri, Ashley Lobo, Nazeer Shaik, Anshul Gandhi,
Zhenhua Liu
IGSC 2023
[pdf]
- Energy-Aware Online Task Offloading and Resource Allocation for Mobile Edge Computing
Yu Liu, Yingling Mao, Xiaojun Shang, Zhenhua Liu, Yuanyuan Yang
ICDCS 2023
[pdf]
- Joint Task Offloading and Resource Allocation in Heterogeneous Edge Environments
Yu Liu, Yingling Mao, Zhenhua Liu, Fan Ye, Yuanyuan Yang
INFOCOM 2023
[pdf]
- Predicting Network Buffer Capacity for BBR Fairness
Umit Akgun, Santiago Vargas, Michael Arkhangelskiy, Andrew Burford, Michael McNeill, Aruna Balasubramanian, Anshul Gandhi, Erez Zadok
ML for Systems 2022
[pdf]
- SLO-Aware Space-Time GPU Sharing for DL Workloads
Ubaid Ullah Hafeez, Anshul Gandhi
Poster @ SOCC 2022
[pdf]
- Metrics for Sustainability in Data Centers.
Anshul Gandhi, Kanad Ghose, Kartik Gopalan, Syed Hussain, Dongyoon Lee, David Liu, Zhenhua Liu, Patrick McDaniel, Shuai Mu, Erez Zadok
HotCarbon 2022 [pdf]