1 / 52

Two Systems to Fundamentally Restructure Mobile Wireless Services

Two Systems to Fundamentally Restructure Mobile Wireless Services - Rajiv Chakravorty - Computer Laboratory

twos
Download Presentation

Two Systems to Fundamentally Restructure Mobile Wireless Services

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Two Systems to Restructure Mobile Wireless Services Rajiv Chakravorty Cambridge University Rajiv.Chakravorty@cl.cam.ac.uk

  2. Mobiles vs PCs as Primary Digital Device 4 B Mobile Penetration 3 B 2 B PC Penetration 1 B 1995 2005 2000 2010 Paradigm shift with Mobile Computing Paradigm shift with Mobile Computing

  3. India/China Calling… Where do you think 4+ Billion mobile users would be?

  4. Two new directions • How to enable an ‘open’ mobile ecosystem? MoB System Architecture User reputation and financial incentive mgmt. Application evaluation    • How to restructure voice-centric mobile networks to support Video? MobiStream System Virtual Channels Perceptual Slice-structured Video   

  5. BUY-1-GET-1-FREE Mobile Operators Inflexible contracts. (SIM)-Locked devices. Inflexible contracts. (SIM)-Locked devices. Coverage not ubiquitous. Coverage not ubiquitous.

  6. Today’s (Closed) Mobile Architecture Applications and mobile devices aggressively controlled by cellular operators VZW VZW VZW VZW Verizon AT&T Yahoo Yahoo Yahoo Yahoo Sprint google google google google No (fine-grained) Competition. No Choice. No (fine-grained) Competition. No Choice. Lack of Innovation. Lack of Innovation.

  7. Why ‘Open’ Closed Networks? Foster Innovation. Respect Users’ Choice. Foster Innovation. Respect Users’ Choice. Enable (fine-grained) Competition. Enable (fine-grained) Competition.

  8. Fundamental Design Principles

  9. ACT 1996 Restructured US telecom Recall that restructuring enabled fine-grained Recall that restructuring enabled fine-grained competition for telecom products & services competition for telecom products & services

  10. Today’s (Closed) Architecture VZW VZW VZW VZW Verizon AT&T Yahoo Yahoo Yahoo Yahoo Sprint ebay ebay ebay ebay No User Choice. Lack of Innovation. No User Choice. Lack of Innovation. No (fine-grained) Competition. No (fine-grained) Competition.

  11. Tomorrow's Open Mobile Architecture Mobile Bazaar (MoB) Bandwidth Location AT&T Bandwidth Shared Web Cache P2P File Verizon Mobile devices can independently create Mobile devices can independently create and sell services (aka 'open' market model) and sell services (aka 'open' market model)

  12. Scenario #1: Bandwidth Aggregation Mobile Bazaar (MoB) Bandwidth AT&T Bandwidth Verizon Enables distributed wireless diversity decisions Enables distributed wireless diversity decisions with mobile users/devices with mobile users/devices

  13. Scenario #2: P2P file transfer service Mobile Bazaar (MoB) p2p file transfer p2p file transfer Enables flexible customization of services based Enables flexible customization of services based on user or app needs (disconnected mode) on user or app needs (disconnected mode)

  14. Scenario #3: Opportunistic Services Mobile Bazaar (MoB) Location information Enables opportunistic time-scale service Enables opportunistic time-scale service interactions (disconnected mode) interactions (disconnected mode)

  15. The MoB Advantage • Better Performance thru Wireless Diversity  Network Diversity  Channel Diversity  Technology Diversity • Financial incentive for Collaborations  Enables flexible, application-level service trades • Flexible customization of services  Users choose services based on price, signal, location, etc.

  16. What are MoB environments? Metros Shops/Malls Buses/Trains Conferences

  17. MoB Contact time is Variable 80 70 60 ~40% stay more than 20 min % of Users 50 40 Observer-based Opportunistic Contacts 30 Signup-based 20 10 0 0 2 1 10 20 50 5 100 200 Duration (in minutes)

  18. Application-layer Services in MoB •Hop-by-hop “pair-wise” service interactions Location A B Shared Bandwidth Cache C How can A be made to trust B, or trust C? How can A be made to trust B, or trust C?

  19. Vito – Building trust thru Reputation •Vito modeled on ebay • 3rd Party Internet service • 4 million transactions/day • Misuse rate very low

  20. Step 1 – Devices register using key-pairs Vito Servers (Online Reputation System) Register Request Register Request Trader (B) Customer (A) Rb,t Ra,t Devices receive signed time-stamp reputation cert. Devices receive signed time-stamp reputation cert. (online registration with public key) (online registration with public key)

  21. Step 2 – Service Interaction and Payment Customer (A) Trader (B) SLP Request (Rep. Cert.-A) SLP Response (Price, Rep. Cert.-B) Token (A, B, Price, t) Service Interactions Customer makes payment first to receive service Customer makes payment first to receive service (Disconnected — No Vito during interaction) (Disconnected — No Vito during interaction)

  22. Step 3 – Token Encash and Feedback Vito Servers (Online Reputation System) Token Encash Feedback for B Token encashed Trader (B) Customer (A) Feedback for A Token encash is implicit positive feedback of B. Token encash is implicit positive feedback of B. Customer provides feedback only if dissatisfied Customer provides feedback only if dissatisfied

  23. Step 4 – Reputation Updates at Night Vito Servers (Online Reputation System) Trader (B) Customer (A) Rb,t Ra,t A small transaction fee is the incentive for Vito A small transaction fee is the incentive for Vito

  24. Remote file download example Microsoft Web Server 3G-1xRTT 3G-EvDO MoB middleware installed in mobile devices C-2 C-1 MoB Middleware Per-connection striping FTP .. Media Streaming Media Streaming FTP WWW WWW Objective: Download Microsoft Vista Logo Image Connection Mgr Rep Mgr. SLP Bluetooth/ WLAN Stripe Mgr Cache Mgr Discovery Layer

  25. P2P file transfer (testbed) 900 800 Download time (seconds) 700 2 Bluetooth Traders w/churn (5MB File) 600 500 High Churn 400 Medium Churn 300 Low Churn 200 100 0 0 100 250 200 50 150 Block size (KB)

  26. Collaborative Web Browsing (testbed) 90 3G-1xRTT (cache not shared) 3G-EvDO (cache not shared) 75 EvDO + 1xRTT (cache not shared) Download Time (sec) MoB (w/ churn + shared cache) 60 MoB (w/o churn + shared cache) 45% 45 38% 40% 41% 30 29% 15 0 ebay s’dot s’forge w’post wisc

  27. Opportunistic location (simulation) 550 500 450 400 1000 Automobiles 2000 Automobiles 5000 Automobiles Location Error (m) 350  12.8 sq. km of mid-town Manhattan  Density of automobiles (1000, 2000, 5000)  “Manhattan random waypoint” mobility model  Constant Speed, but randomly picked [Min, Max]  Adhoc 802.11b radios (80 m range, omni antenna) 100 300 250 200 150 50 2 4 8 10 16 18 0 14 6 12 20 %-tage of GPS-enabled Vehicles 5% GPS nodes ~ 72 m location accuracy 5% GPS nodes ~ 72 m location accuracy

  28. Fundamental Contributions of MoB (ACM Mobicom’05)

  29. Is Mobile Video ready for prime-time? There are challenges… There are challenges…

  30. What is the fundamental issue with 2.5G + 3G Mobile Networks? (Source) (Source) Video (H.263/4) Voice (AMR) (Channel) (Channel) Link + PHY Link + PHY Source-channel optimizedSource-channel Not Optimized Today's Mobile Networks engineered for Voice Today's Mobile Networks engineered for Voice

  31. Mobile Channels are Voice Optimized (e.g., CDMA1xRTT) RLP DATA (L2) f/r-dtch BSC f/r-dtch MUX and QoS Sublayer F-PCDH Control Function BTS PHY Static. Inflexible. No Application Control. Static. Inflexible. No Application Control.

  32. Mobile channels prone to bursty losses RLP = Enabled (stationary) RLP = Disabled (stationary) RLP = Enabled (Mobile)

  33. Streaming Video over Voice Channels Voice/Video/Data Next-generation Voice Next-generation Adaptive Networks Adaptive Networks Static Networks Today's Static Networks Today's Voice Channels Adaptive Channels How to design adaptive mobile networks & apps? How to design adaptive mobile networks & apps?

  34. Our Solution: The “Virtual Channels” Concept Video Streaming (H.264) Applications Partitioned Video Content Application-control for ViCs Rtx = No. of Retransmissions ViC1 ViC2 ViCn EC = Error Correction ViCs Abstraction Link Layer RLP2 RLPn RLP1 MAC Dedicated Shared MAP: PHY Channels App-controlled Channels. Dynamic. Flexible. App-controlled Channels. Dynamic. Flexible.

  35. Dynamic Channel Control – How? Who takes control? Think of a Slider. Apps TCP Voice Video When the slider moves up or down, control of channel shifts between App & Network. Dynamic Control Link+Phy

  36. MobiStream – The BIG Picture Video client Video Proxy Server Video streams Streaming Streaming Push Push App Layer Virtual Channels Virtual Channels Link-Layer Link Layer Servers BTS BSC/NodeB Internet How to intelligently stream video over ViCs? How to intelligently stream video over ViCs?

  37. How to design ‘intelligent’ video apps? 5% loss, 34.6 dB No loss, 44.6 dB 20% loss, 23.7dB 20% loss, 24.1 dB Exploit perceptual value in Video Content Exploit perceptual value in Video Content

  38. Perceptual Slice-structured Video Allzero Interleave ROI Checkboard (slice grp-1) slice grp-2 Percept-SS Specify Region-of-Interest as ‘hints’ in slices Specify Region-of-Interest as ‘hints’ in slices

  39. Perceptual Slicing over ViCs Perceptual Video Codec (ROI segmented frame) ViC-2 (RLP-D) ViC-1 (RLP-E) Slice-2 over ViC-2 Slice-1 over ViC-1 ViCs Abstraction Cellular Link + PHY Subsystem (

  40. Example Frame Partitioning using Motion Vectors Perceptual codecs may segment frames with MVs Perceptual codecs may segment frames with MVs

  41. Experimental results for foreman QCIF video seq. (300 frames) Lossy channel A.FEC + ViCs (uncorrelated channels) A.FEC + MDC + ViCs (uncorrelated channels) raw decode (no loss) Distribution 50% > 1.0 dB 30% > 2.5 dB 10% ~ 4-7 dB PSNR

  42. Sliced Video benefit using ViCs Relative PSNR (dB) Frame No. (GoP Sequence) Up to 4 dB benefit using ViCs (Foreman) Up to 4 dB benefit using ViCs (Foreman)

  43. Perceptual-SS Improves Subjective Picture Quality Checkboard Perceptual-SS

  44. Fundamental Contributions of MobiStream (IEEE Infocom’06) • Virtual Channels  Gives partial control to applications • Perceptual Slice-structured Video  Improves subjective video quality • Fine-grained error resilience  Gracefully overcomes bursty losses using slices • Applications-defined link layer behaviour  Low-overhead Dynamic Channel Management (

  45. Other Contributions  Flow Aggregation (IEEE Infocom’03)  GPRSWeb (ACM Mobisys’03, 20K+ lines of code)  Optimizing Wireless Services (ACM Mobicom’04)  Mobile Access Router (ACM Mobisys’04)  Exploiting diversity for Audio (IEEE Infocom’05)  Wireless Integration and handovers (Percom’04)

  46. Innovative Wireless System Architectures, Applications and Services Future Wireless Network services Spectrum-agile Mobile Radio Networks Mobile Multimedia Communication System Design

  47. Innovative Wireless System Architectures, Applications and Services Next Generation Wireless Networks Spectrum-agile Mobile Networks Mobile Multimedia Communication System Design

  48. Reputation system – Vito Evaluation  500 users (fraction of them malicious)  Services priced randomly (1—5 units )  Price increment/reduction factor (0.01)  End-of-day model (the 80-20 model)  Up to 50 services/day (max. 1000)

  49. Impact of User Selection Policy Services Left after 20 days Selection Policy Good Users Reputation Price Reputation/Price 528 788 859 Malicious Users 73 789 82 Virtual Currency left after 20 days Selection Policy Good Users Reputation Price Reputation/Price 125 98 124 Malicious Users ~ 0.0 107 0.7

  50. Impact of Malicious Users Malicious %tage Services Left (20 days) Good Users Malicious 1% 868 99 5% 870 89 10% 869 94 20% 40% 859 804 82 87 Reputation-driven policies insensitive to Reputation-driven policies insensitive to the fraction of malicious users the fraction of malicious users

More Related