Writing /w/mobilab/wiki/data/cache/5/5d0da8c26a9df7140abc32f48a31b7ef.i failed
Unable to save cache file. Hint: disk full; file permissions; safe_mode setting.
Writing /w/mobilab/wiki/data/cache/5/5d0da8c26a9df7140abc32f48a31b7ef.metadata failed
 
Writing /w/mobilab/wiki/data/cache/5/5d0da8c26a9df7140abc32f48a31b7ef.i failed
Unable to save cache file. Hint: disk full; file permissions; safe_mode setting.
Writing /w/mobilab/wiki/data/cache/5/5d0da8c26a9df7140abc32f48a31b7ef.xhtml failed

Welcome to the MobiLab Wiki!

Overview

MobiLab is a global-scale live laboratory for supporting mobile computing science.

People

Faculty:

Students:

Partners

Collaborators

Research Activities

  • Mobile device state introspection. We have developed a tool with which researchers and developers can determine exactly which hardware-software operations in multithreaded applications and operating systems are on the critical timing paths of user-visible transactions. This tool is called Panappticon, and is described in more detail at the main project website.
  • Connecting mobile devices with their environments. A major research sub-goal is to enable a new class of plug-and-play sensing and data collection peripherals for the mobile phone. To that end, our focus in the first year of this project has been on porting our prior work on the iPod/iPhone-based HiJack (a phone accessory framework) to different phone platforms and operating systems (including Android, Windows Mobile, and Nokia). Activities in these areas include: (i) Characterizing Available Audio Headset Power; (ii) Redesigning HiJack Around Lower Power; and (iii) Communicating over the Mobile Phone's Audio Port. We have also released a firmware package that can be used by researchers to simplify the development of sensing systems containing Arduino BT based sensing nodes.
  • * Application market. TODO…
  • Resource sharing and management. TODO…
  • Computation management. TODO…
  • Real-time data publishing. TODO…
  • Programmability support for user input. TODO…
  • Analyzing common data sets. TODO…

Research Findings

Mobile device state introspection

Reddit News is a popular closed-source application on the Android Market that has millions of downloads. When doing transaction analysis, Panappticon revealed a large number of transactions with latencies longer then 1 second. We found that scheduling preemption is the main reason for this high latency, and that this latency is a result of lengthly writes to nonvolatile storage, Although this problem could be worked around within the application, that would require application developers to have a deep knowledge of subtle operating system/hardware interactions. Instead, it might be better to modify the operating system scheduler or nonvolatile storage semantics. This finding is merely an example of the sorts of findings that developers and researchers can quickly arrived at using the Panappticon tool.

We also found a subtle problem with the Android platform scheduler and interactive power management governor when used with applications that interact with the network or disk. For transactions taking longer than 60 ms, the dynamic voltage and frequency scaling policy dramatically reduces user-perceived performance. The governor increases CPU frequency only when the CPU has recently been heavily loaded. However, low loading doesn't imply that CPU frequency is irrelevant to user experience. When user transactions are long enough to contain network or disk transactions, the CPU necessarily spends some time idle and the operating frequency is reduced. This is true even if it is fully on the critical path during non-network/disk portions of the transaction. The problem can potentially be corrected by treating the CPU as active during network and disk access time on critical paths in user transactions, thereby causing the CPU frequency governor to more precisely base its decisions on whether decreasing CPU frequency will degrade user experience. Again, this finding is another example of the sorts of uses developers and researchers may put Panappticon to.

Connecting mobile devices with their environments

Our major activities and findings include:

Characterizing Available Audio Headset Power. Audio headset-based peripherals need to run from the energy they can harvest from the headset port. Our early results using the iPhone and iPod platforms showed considerable power was avaiable but recent results have shown dramatically less power is available from most other phones. This makes a direct application of the original HiJack system to alternate platforms challenging. We have also explored the microphone bias voltage, available on almost all phones, and found substantially less power, but also less diversity in available power.

Redesigning HiJack Around Lower Power. Given the limited power budget available across most phones, we have rearchitected the HiJack platform to operate in two ways: a purely analog solution that can achive operating power of 200-300 uW and a digital version that uses a more efficient energy harvesting strategy but still requires a peak power approaching a few milliwatts. However, we have been working on duty cycling the overall system to lower power.

Communicating over the Mobile Phone's Audio Port. A goal of this research program is to evaluate a range of different hardware and software implementations of various communications blocks including full receive and transmit chains, and many of their elements. To explore some of the basic issues, and to obtain concrete power, performance, and complexity figures (as well as an intuitive understanding of the tradeoffs), we implemented receive and transmit chains for communicating between a mobile phone and a microcontroller over the phone's headset port. We characterized the signaling and ability of the audio jack, designed circuits and software to transfer data, and evaluated the performance of our designs.

Communications Viability. We find that bi-directional communications over the mobile phone's audio headset port is viable using a binary FSK channel at a data rate of 8.82 kbps. We also find orders of magnitude difference in power for a functionally similar decoder running as a software-defined receiver on the mobile phone (implementing a non-coherent FSK demodulator) vs the same functions that are implemented as a hardware-accelerated receiver with bit-level timing extraction using timer compare, timer capture, and hardware UART peripherals. A range of experiments allowed us to find distinct power, performance, error rates, and algorithm complexity points in the design space.

Power Limitations. Our early results using the iPhone and iPod platforms showed considerable power was available – roughly 16 mW – from the audio headset port. Recent results have shown dramatically less power is available from most other phones: HTC Droid (6.8 mW), Google Nexus One (3.0 mW), HTC Tilt 2 (2.7 mW), Samsung Focus (2.4 mW), and even less from low-end feature phones. Power differences exceeding a factor of seven exist across the phones we tested. Meanwhile, the open-circuit voltage and short-circuit current are quite different as well. This makes a direct application of the original HiJack system to alternate platforms challenging. We have also explored the microphone bias voltage, available on almost all phones, and found substantially less power, but also less diversity in available power: iPhone 4 (0.81 mW, Nokia N900 (0.63 mW), Nokia E71 (0.59 mW), Samsung Focus (0.58 mW), Google Nexus One (0.53 mW), HTC Droid (0.46 mW), Nokia N3600 (0.41 mW), HTC Tilt 2 (0.29 mW).

Application market

TODO…

Resource sharing and management

TODO…

Computation management

TODO…

Real-time data publishing

TODO…

Programmability support for user input

TODO…

Analyzing common data set

TODO…

Publications

Citation counts are taken from Google Scholar.

  1. Lide Zhang, David R. Bild, Robert P. Dick, Z. Morley Mao, and Peter Dinda, “Panappticon: Event-based Tracing to Optimize Mobile Application and Platform Performance,” under review, International Conference on Mobile Systems, Applications, and Services, June 2013.
  2. Sonal Verma, Andrew Robinson, and Prabal Dutta, “AudioDAQ: Turning the Mobile Phone’s Headset Port into a Universal Data Acquisition Interface,” in Sensys'12: Proceedings of the Tenth ACM Conference on Embedded Networked Sensing Systems, Nov. 2012.
  3. L. Zhang, M. S. Gordon, R. P. Dick, Z. M. Mao, P. Dinda, and L. Yang, “ADEL: An automatic detector of energy leaks for smartphone applications,” in Proc. Int. Conf. Hardware/Software Codesign and System Synthesis, Oct. 2012.
  4. Andrew Robinson, Sonal Verma, and Prabal Dutta, “Demo: AudioDAQ: Turning the Mobile Phone’s Headset Port into a Universal Data Acquisition Interface,” In IPSN'12: Proceedings of the Eleventh International Conference on Information Processing in Sensor Networks, Apr. 2012.
  5. Ye-Sheng Kuo, Sonal Verma, Thomas Schmid, and Prabal Dutta. “Hijacking Power and Bandwidth from the Mobile Phone's Audio Interface,” In DEV'10: Proceedings of the First Annual Symposium on Computing for Development, London, United Kingdom, Dec. 17-18, 2010.
  6. L. Zhang, B. Tiwana, Z. Qian, Z. Wang, R. P. Dick, Z. M. Mao, and L. Yang, “Accurate online power estimation and automatic battery behavior based power model generation for smartphones,” in Proc. Int. Conf. Hardware/Software Codesign and System Synthesis, Oct. 2010, pp. 105–114. Cited by 142 papers.

Links

Broader Impacts

  • HiJack, our hardeware/software toolkit for connecting mobile phones to external sensors for sporadic/interactive sensing applications, is available through Seeed Studios. The available artifacts include the HiJack Main Board, a Development Pack, and a Programmer. As of June 2012, approximately 375 HiJack Development Kits were in use in over 35 countries and on 6 continents by academic, industrial, and hobbyist users.
  • techBASIC from ByteWorks has integrated support for HiJack, enabling students, hobbyists, educators, and scientists alike to do scientific programming on iPods, iPhones, and iPads that interact with externals sensors and other devices. The techBASIC app is available on iTunes.
  • MobiPerf and Powertutor have been used by researchers from more than 18 universities (including University of San Francisco, U.S., University of Washington, U.S., CMU, U.S., Lehigh University, U.S., New Jersey Institue of Technology, U.S., Mobile Technologies Research group, Concordia University, Canada, University College London, University of Trento, Italy, Singapore Management University, SG, University of Moratuwa, Sri Lanka, National Institute of Technology, Inda, Ghent University, Belgium, Xiamen University, China, Peking University, China, Seoul National University, Korea, Chungnam National University, Korea, Technical University of Darmstadt, Germany, Department of Informatics and Telecommunications, University of Athens, Greece) and 5 companies (including Bay Web Soft, U.S., Motorola, U.S., Intel, U.S., Green code lab, France, Samsung, Korea). These tools have been actively used by many smartphone users, exceeding more than 100,000 runs for MobiPerf and 100,000 downloads for PowerTutor.
  • Refer to this link to understand the geographic coverage of the different users running MobiPerf software.

Task status

start.txt · Last modified: 2013/01/21 13:16 by dickrp
 
Except where otherwise noted, content on this wiki is licensed under the following license: CC Attribution-Share Alike 3.0 Unported
Recent changes RSS feed Donate Powered by PHP Valid XHTML 1.0 Valid CSS Driven by DokuWiki