Sensing and Energy Research Group (SERG)

Sensing and energy are emerging cross-cutting physical concerns in computer systems. The proliferation of embedded and personal devices such as networked sensors and mobile phones gives computer systems increasing capability of gathering data from and adapting to the physical world and personal activities. Energy constraints at different scales can affect battery life and thus user experiences in embedded devices, and can be a limiting factor for the growth of IT infrastructure. Capturing, understanding, and exploiting sensor data and energy characteristic will continue to revolutionize how technologies can serve people, provide awareness, improve efficiency, and reduce resource consumption.

SERG at MSR conducts fundamental and systems research on the interface and interactions between computing systems and the physical world. We develop new architecture, hardware and software platforms, system and network principles, energy and data management strategies, signal and information processing techniques, and novel applications to advance the state of art in sensing and energy-efficient systems. The application domains of our research include wireless sensor network,  mobile systems, wearable computing, personalized on-line services, data centers, and green cloud computing.

Projects:

  • Mobile Subjective Sensing

The essence of the project is to sense mobile phone users as subjective individuals, rather than "objects" in motion. Leveraging sensors on mobile devices and in the environment, ubiquitous network connectivity, and increasingly powerful computing cloud, we are developing a framework to continuously sense the mobile users and to extract subjective information such as user presence, preference, and priority. Our research efforts, code named Munich (Mobile Users in Non-Intrusive Computing Hierarchies) include

  • LittleRock and SensoryPhone: A heterogeneous multiple processor (core) architecture and software stack for mobile devices to support low power continuous sensing. 
  • CondOS: abstrations and a run-time system for mobile context aware operating systems and applications;
  • Everest: subjective data management across computing hierarchies;
  • BingNow!: a subjective mobile search engine;
  • Pocket Cloudlets: cloud services, such as search or content browsing, subjectively pre-fetched and cached on mobile devices.  
  • Location Sensing 

Although location sensing is a well studied topic, we continue to make breakthroughs by leveraging new sensing modalities and new processing paradigms. For in-door location sensing, we investigated using public FM-radio signals as signtures, which by combining with WiFi signatures gives 99%+ room-level location accuracy. For out-door location sensing, we found a way to offload GPS signal processing to the cloud and thus reduce the device side GPS sensing energy consumption by over 99%. Location sensing is an integral part of the CLEO project to enable low cost and energy efficient participatory sensing.

  • Wearable Computing

We research on novel wearable devices that can help human monitor themselves as well as interact with others and the environment. Our foci are on energy scavenging, energy management, low power communication, and sensing.

Data centers are complex physical systems hosting massive online and cloud services. Some mega-scale data centers can contain hundreds of thousands of servers and consume over a hundred mega-Watts of power. Improving their efficiencies is both an important cost reduction measure for online service companies and a part of the social responsibility of the IT industry as a whole.

The goal of the project is to sense and understand key operating parameters for data centers and to ultimately improve their operation efficiency. Our research efforts and achievements include: MeshID: tracking the physical assets in data centers; Genomotes: wireless sensor nodes for dense temperature and humidity sensing; RACNet: a reliable data aquisition network; Cypress: a compressive data management system for streaming sensor data; Prophet: a data-driven capacity management tool for data centers; CoolShift: a thermo-aware modeling tool for data centers as a cyber-physical system. 

Addressing challenges in reducing enterprise and cloud computing energy consumption, we design and develop tools, analysis methodologies, and control systems to measure, model, and optimize computer systems for energy efficiency.

Research efforts include JouleMeter: a tool that provides visibility into computer, virtual machine (VM), and application level power consumption using software observable performance counters; Cuanta: a methodology and tool for quantifying VM performance interference in consolidated servers, especially due to on-chip resource sharing; VPS: Power budgeting for virtualized infrastructures.  

Internal Research Collaborations:

Past Projects:

  FlashDB | mPlatform | SenseWeb | MSR Sense | SONGS |Tinker

Publications

    2014

    2013

    2012

    2011

      2010

      2009

      2008

      2007

      2006

      2005

      Highlights:

      People
      Michel Goraczko
      Michel Goraczko

      Aman Kansal
      Aman Kansal

      Jie Liu
      Jie Liu

      Suman Nath
      Suman Nath

      Bodhi Priyantha
      Bodhi Priyantha

      Mohammed Shoaib
      Mohammed Shoaib

      Di Wang
      Di Wang

      Visiting Researchers/Collaborators:

      Interns/Student Consultants:

      2013: Vageesh Devarayasamudra Chandramouli, Jeremy Gummeson, Yanlin Li, Bin Liu, Yixin Luo, Azalia Mirhoseini, Shahriar Nirjon, Alejandro Eden Reyes, Ana Riekstin, Alexandra Vtyurina, Di Wang, He Wang, Alexandra Vtyurina 

      2012: Woosuk Lee, Felix Xiaozhu Lin, Justin Lu, Prashanth Mohan, Moaj Musthag, Amanda De Paula, Demostenes Zegarra Rodriguez, Abu Sayeed Saifullah, Trang Thai, Di Wang, He Wang,

      2011: Hossein Ahmadi, Xuan Bao, Arka Bhattacharya, Yin Chen, Cristina Dominicini, Yuanhua Lv, Radhika Mittal, Moo-Ryong Ra, Alan Roytman, Negin Salejegheh, Trang Thai, Tingxin Yan, Tao Zhang, Ziguo Zhong.

      2010: Shah Amini, Christina Delimitrou, Michaela Goetz, Oliver Kennedy, Pavan Kumar, Hong Lu, Miguel Palomera Perez, Aveek Purohit, Heitor Ramos, Peixiang Zhao