Current Projects

The projects below have been generously supported by the National Science Foundation, the state of Massachusetts, VMware, Intel, Google, and Amazon. Most projects are collaborations with other UMass Amherst faculty, undergraduate and graduate students, and other external collaborators from industry and academia.

While the cloud’s energy demand grew more slowly than expected over the past decade due to aggressive energy-efficiency optimizations, there are few remaining optimization opportunities using traditional methods. As a result, continuing the cloud's exponential growth will translate into exponentially rising energy demand, which will position it as one of the primary contributors to global carbon emissions. To address the problem, this project elevates carbon to a first-class metric in designing a sustainable and reliable cloud-edge software infrastructure that can enable continued exponential growth. This project is funded by the National Science Foundation and VMware under grant CNS-2105494.
The rapid transition to an energy system based on renewable electricity sources instead of fossil fuels is one of the great challenges facing humanity. It also offers a unique opportunity to reshape our society to be more equitable and more sustainable. To address this challenge, ELEVATE draws upon expertise in a wide range of research fields to create an innovative graduate STEM program focused on resilience and equity in the energy transition. ELEVATE is funded by the National Science Foundation NSF Research Traineeship (NRT) program grant DGE-2021693 and Growing Convergence Research (GCR) program grant SES-2020888.

Optimizing Distributed Machine Learning for Transient Servers

Large-scale distributed machine learning (ML) is expensive to run on cloud platforms. To reduce cost, cloud platforms now offer cheap transient servers, which they may revoke at any time. This project is re-designing traditional distributed ML algorithms to use looser forms of synchrony, and designing adaptive policies for selecting transient servers based on their performance, cost, and volatility. This re-design will lower the cost of running distributed ML by enabling them to use cheap transient servers. This project is funded by the National Science Foundation under grant CNS-1908536.
The NSF “Open Cloud Testbed” (OCT) project is building and supporting a testbed for research and experimentation into new cloud platforms – the underlying software which provides cloud services to applications. Testbeds such as OCT are critical for enabling research into new cloud technologies – research that requires experiments which potentially change the operation of the cloud itself. This project is funded by the National Science Foundation under grant CNS-1925464.

Past Projects

Software-defined Solar-powered Systems

This project focuses on the design of "smart" software-defined solar (SDS) systems that self-regulate the power they let "flow" into the grid. Similar to data transmission in the Internet, SDS systems dynamically regulate the solar energy that flows into the grid to maximize its available solar capacity, maintain a balanced supply and demand, and fairly share the available capacity among connected users. The project is funded by the National Science Foundation under grant CNS-1645952.

Managing Risk in Cloud Platforms

Cloud computing has become the foundation of our information-based economy, providing the computing power necessary for advances in all sectors. However, cloud platforms expose users to a number of hidden risks. This project develops techniques for efficiently managing risk in cloud platforms by adapting and extending concepts from economics. A key goal of the project is to elevate risk management to a first-class systems design principle. The project is funded by the National Science Foundation under grant CNS-1802523, Google Faculty Research Award, and Amazon in the form of cloud credits.
This project developed a programmable data-driven testbed to enable a range of research and education activities focused on sustainable smart buildings. Our testbed is comprised of two components: a data collection component based on a large number of real instrumented buildings of various types to gather anonymous fine-grained energy usage and operational data at scale; and a programmable smart building component that uses programmable power sources and electrical loads to enable the evaluation of a variety of energy management mechanism. This project was funded by the National Science Foundation under grant CNS-1405826.

Utility-driven Smart Energy Services

This project developed utility-driven smart energy services to improve grid energy efficiency, encourage energy conservation, and promote local renewable energy sources, such as rooftop solar. The research focused on developing new energy analytic techniques for smart meter data, combining these in novel ways to create specific applications that improve energy efficiency, packaging those applications as cloud-based web services that are accessible to end users, and evaluating their impact on user behavior and energy efficiency. This project was funded by the National Science Foundation under grant CNS-19254646 and a grant from the Massachusetts Department of Energy Resources (DOER).

System Support for Transient Cloud Computing

In many emerging scenarios, clouds and data centers can only provide transient server availability. This project was the first to develop system support for transience that treated it as a first-class design principle. Since traditional fault-tolerance techniques are too expensive for handling transient server availability, the project developed new transience-aware mechanisms that balanced cost, performance, and availability. This project was funded by the National Science Foundation under grant CNS-1422245, a 2016 Google Faculty Research Award, and Amazon (in the form of cloud credits).

Enhancing Privacy in Smart Buildings

The design of smart buildings that automatically optimize energy usage is important for improving energy-efficiency. Unfortunately, a key barrier to the broad adoption of energy-efficiency optimizations is that Internet-connected sensors often leak private information about user behavior. This project designed low-cost, non-intrusive, privacy-enhancing techniques that reduced the private information leaked through smart devices, while still permitting the sophisticated analytics, control, and verification necessary to enable energy optimizations for smart buildings. This project was funded by the National Science Foundation and Intel under grant CNS-1505422.
This project developed BenchLab, an open, flexible community infrastructure comprising applications, workloads and tools to enable realistic performance evaluation and benchmarking by systems researchers. BenchLab is an open framework where source code and workload datasets are freely available for modification and use by researchers for their specific experiments. The framework consists of a suite of server-side benchmark applications and workloads that represent cloud, mobile web, and green computing environments. This project was funded by the National Science Foundation under grant OAC-1339839.
The project developed a model-based paradigm to design automated load discovery, monitoring, and scheduling techniques that are low in cost, highly transparent, and privacy preserving. The project bootstrapped an online data repository to enable new techniques for modeling loads based on their energy usage. The project leveraged these models to automatically discover the set of loads a building operates, and monitor their electricity usage using only aggregate electricity recorded by a building smart meter. This project was funded by the National Science Foundation under a NSF CAREER grant CNS-1253063.