Outsourcing Decisions in Mobile Image Processing Oslo 27.04.2012 - PowerPoint PPT Presentation

Power and Latency Impacts of Outsourcing Decisions in Mobile Image Processing Oslo 27.04.2012

Outline 5/2/2012  Team members  Introduction  Application description  Test Environment  Measurements and results  Analysis of the results  Outsourcing decision making algorithm  Future work  Conclusion 2

Team Memebers 5/2/2012  Ali Zaher and Ali Ahmad  Niklas Dürr and Nicolas Oliver Stamer Department of Informatics  Oslo University School of Business Informatics   and Mathematics P.o.Box 1080, Blindern  University of Mannheim  NO-0316 Oslo, Norway  A5, 6  Tel: +47 228 45581  68159 Mannheim, Germany  Email: alizah@ifi.uio.no  Email: nduerr@mail.uni-  Email: aliaah@ifi.uio.no  mannheim.de Email: nistamer@mail.uni-  mannheim.de 3

Introduction 5/2/2012  Early days of mobile phones: Voice and then sms.  Current days: Data Traffic (Video, images, emails,…)  data traffic has taken over voice traffic on mobile networks already in 2010. 4

Introduction 5/2/2012  The world most selling phone, Nokia 1100. 2 weeks standby time.  The world most selling phone, Nokia 1100, 2 weeks standby time.  2012 mobile phones with Quad core at 1.5 GHz, battery 1800 mAh,  2012 mobile phones with Quad core at 1.5 GHz, battery 1800 mAh, connectivity: Wi-Fi: IEEE 802.11 a/b/g/n, HSDPA 21 Mbps connectivity: Wi-Fi: IEEE 802.11 a/b/g/n, HSDPA 21 Mbps  2003 mobile phones with CPU ARM-9 104 MHz, battery 850 mAh,  2003 mobile phones with CPU ARM-9 104 MHz, battery 850 mAh, connectivity: GSM 24 - 36 kbps connectivity: GSM 24 - 36 kbps 5

Introduction  How do mobile phone batteries follow related 5/2/2012 to Moore’s Law? 6

Introduction 5/2/2012  What about sourcing out the power hungry apps to the cloud?  More power efficient??  Faster execution???  “Make or buy” decision from economics  Image processing algorithm, why?  April 23: Facebook offers 23 million shares and $300 million in cash to Instagram (almost 1B$) 7

Application desciption- App on Android 5/2/2012  Mobile device: "HTC Desire S"  CPU frequency: 1,0 GHz  RAM: 768 MB   Server: 2,4 GHz Unix-based server  RAM:4 GB.  8

Test Environment 5/2/2012  The remote execution is implemented with Java Sockets  Open source image manipulating algorithms of JH Labs  A relatively big image with 600x300 pixels and 57 KB and a small image with 400x200 pixels and 29 KB. 9

Test Environment - Algorithm selection 5/2/2012 10

Measurements and Results- Total Energy 5/2/2012 11

Measurements and Results- Energy divided 5/2/2012 12

Measurements and Results- Energy with compression 5/2/2012 13

Measurements and Results- Latency 5/2/2012 14

Measurements and Results- Latency in 3G vs WiFi 5/2/2012 15

Measurements and Results- Power and Latency in WiFi with compression 5/2/2012 16

Measurements and Results- Power and Latency in 3G with compression 5/2/2012 17

Outsourcing decision making algorithm 5/2/2012  Signal Strength is a factor in 3G.  Other factors: current bandwidth available, number of users connected to the same base station.  Log for every execution:  The image size in bytes  The image algorithm name.  Executing locally or on the server.  Execution time  Signal strength  Connection type (whether 3G or WiFi)  Transmission time Update the log file gradually to keep it simple. For similar Signal  Strength entries, apply: 18

Outsourcing decision making algorithm- Example 5/2/2012 Logged by the phone Processed log Signal Transmission Image no. of Signal Transmission Image no. of Strength time size transmissions Strength time size transmissions -96 20.583 1000000 6 -65.4 11.5 600000 1 -91 58.826 1000000 2 -90.2 28.222 705821 1 -90 39.985 1000000 2 -70.6 12 304581 1 -85 39.398 1000000 1 -80.8 18.0003 242456 1 -81 74.242 1000000 2 -65.2 13.222 705821 1 -80 74.242 1000000 1 -71.6 12 304581 1 -72 39.398 1000000 1 -80.0 18.0003 242456 1 -71 39.398 1000000 1 -95.5 11.5 600000 1 -70 39.398 1000000 1 -91.2 28.222 705821 1 -65 21.967 1000000 12 -70.2 12 304581 1 -80.8 18.0003 242456 1 -90.2 28.222 705821 1 -85.4 12 304581 1 -90.8 18.0003 242456 1 -65.4 22.5 1000000 10 -95.8 12.5 600000 5 19

Outsourcing decision making algorithm 5/2/2012 1) Extract the needed information as log in before. 2) If the algorithm is not complex, then execute locally and log as described before. 3) In case the algorithm is somehow complicated, we check for the expected transmission time at the current signal strength and compare it to the recorded execution time locally. If it is smaller, then we execute on the server and wait for the server result to log. 4) In case the expected transmission time is greater than the local execution time, then execute locally and log as described before. 5) In case the algorithm is complicated, then we check if the phone is in power saving mode. If it is not, then we execute on the server and wait for the server result to log. 6) If the phone is in power saving mode, and the expected transmission time at the current signal strength is smaller than local execution time, the user can decide to quit the operation or to outsource the operation to the server 20

Outsourcing decision making algorithm 5/2/2012 21

Future work 5/2/2012 1) Apply the outsourcing decision making algorithm on large data 2) Check for 4G 3) Go for bigger image sizes 4) Add more complicated image filters 5) Look at video algorithms 22

Conclusion 5/2/2012 23

REFERENCES 5/2/2012 [1] Commons, John Rogers. 1931. "Institutional Economics", American Economic Review,  Vol. 21, pp. 648-657 [2] Williamson, Oliver E. 1981. "The Economics of Organization: The Transaction Cost  Approach", The American Journal of Sociology, 87(3), pp. 548-577 [3] Byung-Gon Chun, Sunghwan Ihm, Petros Maniatis, Mayur Naik, and Ashwin Patti.  2011. CloneCloud: elastic execution between mobile device and cloud. In Proceedings of the sixth conference on Computer systems (EuroSys ’11). ACM, New York, NY, USA, 301 - 314. [4] R. Rana, C.T. Chou, S. Kanhere, N. Bulusu and W. Hu, "Ear-Phone: An End-to-End  Participatory Urban Noise Mapping System", in Proceedings of IPSN’10, April 2010. [5] Ahmed A. Abukmail and Abdelsalam (sumi) Helal. 2007. Energy Management for  Mobile Devices through Computation Outsourcing within Pervasive Smart paces.Submitted to the IEEE Transactions on Mobile Computing [6] Mei, C., et al., Dynamic Outsourcing Mobile Computation to the Cloud. 2011,  Department of Computer Science and Engineering, University of Minnesota: Twin Cities. [7] R. Kemp, N. Palmer, T. Kielmann, and H. Bal. Cuckoo: a Computation Of floading  Framework for Smartphones. In MobiCASE ’10: Proceedings of The Second International Conference on Mobile Computing, Applications, and Services, 2010. [8] Android Interface Definition Language,  http://developer.android.com/guide/developing/tools/aidl.html 24

REFERENCES 5/2/2012 [9] Bernd Girod and Vijay Chandrasekhar Stanford University, Radek Grzeszczuk  Nokia Research Center and Yuriy A. Reznik Qualcomm 2011, Mobile Visual Search: Architectures, Technologies, and the Emerging MPEG Standard [10] Nister, D.; Stewenius, H.; , "Scalable Recognition with a Vocabulary Tree,"  Computer Vision and Pattern Recognition, 2006 IEEE Computer Society Conference on , vol.2, no., pp. 2161- 2168, 2006 [11] http://www.jhlabs.com/ip/filters/index.html  [12] http://instagr.am/  [13] http://powertutor.org/  25

Q&A 5/2/2012 Thanks for your attention 26

Extra slide 5/2/2012 27