Network Geointelligencemahbub/snpd_keynote.pdf · See “performance comparison of 3G and metro-scale wifi for vehicular network access”, ACM International Measurement Conference

Keynote Speech, ACIS SNPD 2011, Sydney, 6 July 2011

Network Geointelligence Coping Bandwidth Uncertainty in High-Speed Mobility

MAHBUB HASSAN Professor – Computer Science and Engineering

University of New South Wales, Sydney, Australia

Acknowledgements

Dr. Salil Kanhere (my colleague at UNSW) Mr. Jun Yao (former PhD student, now with Freelancer.com)

Two Amazing Developments in Mobile Computing

2000: 9kbps (GPRS) 2008: 3.6 Mbps (HSDPA) 2010: 21 Mbps (HSPA)

Exceeded 1GHz processing speed Ultra fast and high capacity memory

Sharp Increase in Mobile Network Speed is Creating New Market Opportunities

Vehicular Broadband

Smart phones can enjoy streaming!

Heading for a Seamless Mobile Internet

(Fixed) Internet

mobile access is surpassing fixed access

mobile access to Internet

Bandwidth Challenges for Mobile Internet

  Peak capacity is not a challenge anymore   Telstra already announced 42 Mbps peak rate

  But wireless bandwidth is extremely uncertain   Actual available bandwidth can vary anywhere from 0 to peak rate

  High bandwidth uncertainty hinders reliable and quality commercial services   Optimal delivery critically depends on knowledge of bandwidth

Quiz: if peak capacity increases, would uncertainty increase or decrease?

Presentation Overview

  Geo-sensitivity of mobile bandwidth   Geointelligence (to reduce bandwidth uncertainty)   Applications of geointelligence

  Mobile streaming   Mobile multihoming

  Conclusion and on-going work

Is mobile bandwidth geo-sensitive?

If so, what are the implications?

It seems that location determines your luck with bandwidth

  Wired.com’s iPhone 3G survey in 2008 (www.wired.com)

3G bandwidth varied significantly at country and region level

High-speed mobility escalates bandwidth uncertainty

Taken from “HSDPA performance and evolution”, Ericsson Review, No. 3, 2006

Stationary Driving

UNSW MEASUREMENT CAMPAIGN IN 2008

It seems 3G bandwidth is geo-sensitive. Can we quantify this geo-sensitiveness?

J. Yao, S. Kanhere and M. Hassan, "An Empirical Study of Bandwidth Predictability in Mobile Computing", WiNTECH’08 (in ACM MOBICOM 2008), San Francisco, 19 Sep 2008.

Measurement Architecture bandwidth is measured every 200 meters of a road

Internet

Probe Server @ UNSW

Probe Client

Provider C (pre-wimax)

Provider A (HSDPA)

Provider B (HSDPA)

Downlink Probe (Packet Train)

Probe Trigger (every 200m)

Measurement Hardware/Software   Off-the-shelf Hardware (Soekris)   Totally user-driven (no support from service provider)

Routes and Trips   Inbound 7Km & Outbound 16.5Km (total 23.5 Km) at 70-80 Kmph   75 repeated trips (Aug’07 – Apr’08)

inbound

outbound

UNSW

Quantizing the bandwidth “signal”

Probability distribution of different locations location = 500 meter of road

They can be very different! Bandwidth is indeed geo-sensitive.

31 19

Differences in bandwidth distributions between adjacent road segments (L1 distance values are within 0-2)

Mobile apps could be in for a bumpy ride!

Bandwidth Varies Significantly at Many Geographical Scales (individual trip data for 3 trips - inbound)

Data for Provider C (pre-WIMAX)

Bandwidth Varies Significantly at Many Geographical Scales (average bandwidth from 75 trips)

3G bandwidth exhibits significant geo-sensitivity

A CASE FOR GEOINTELLIGENCE

Bandwidth Entropy Quantifying Bandwidth Uncertainty with Information Theory

  Entropy quantifies uncertainty in data

  Lower the entropy, lower the uncertainty, better the predictability

  Entropy=0 completely deterministic   Entropy=log2X completely random

Example of a random variable with 2 possible outcomes, 0,1.

When X is a completely random process

Location-based analysis reduces bandwidth uncertainty (case for geointelligence)

How Geointelligence Can Help (assume it stores average bandwidth observed in a given location from the previous trips)

At the entry to location #7, geointelligence would give 133 kbps, but a link monitor agent would give 544

Convergence to 68 would be faster and smoother if started from 133 instead of 544

Avg(114,153) = 133

Root Mean Square Error Comparison (averaged over all 75 trips)

Error with link monitor (no geointelligence)

Error with geointelligence

MOBILE STREAMING

J. Yao, S. Kanhere, and M. Hassan, "Quality Improvement of Mobile Video Using Geo-intelligent Rate Adaptation", IEEE WCNC 2010, Sydney, 18 April 2010.

Adaptive Video Streaming

  Store several streams of different quality (PSNR) for the same video   Current bandwidth is continuously monitored   Switch streams (quality or PSNR) according to current bandwidth   Adaptation algorithms – TFRC, 3GPP PSS, HTTP, proprietary,

TFRC – TCP Friendly Rate Control

  A widely discussed algorithm for UDP-based adaptive multimedia

  TCP-like AIMD (additive increase multiplicative decrease) congestion control   Slow ramp up for sudden low to high bandwidth

(wastes high PSNR opportunities)   Packet loss for sudden high to low bandwidth

(quality may degrade beyond acceptable level)

Geo-TFRC (TFRC with access to geointelligence)

  Goal: To help TFRC adapt to sudden bandwidth variations at location crossings

TFRC and Geo-TFRC Simulation

Foreman.qcif self-concatenated to create a 30 min video lasting the entire trip

geointelligence

Video Rate Adaptation Evaluation in ns-2 Based on Evalvid-RA (Lei et al. ’07) , Evalvid (Ke et al. ’08)

Video quality measurement (video quality is affected by packet loss from buffer overflow at cellular tower)   The PSNR metric

  For acceptable video quality: PSNR >= 31 (viewing is ‘disrupted’ for low PSNR)

PSNR Comparison (cont.) Geo-TFRC

TFRC

Fraction of Time With Poor Streaming Quality (PSNR < 31)

50% more disruptions

500% more disruptions

MULTIHOMING

J. Yao, S. S. Kanhere, M. Hassan, "Geo-intelligent Traffic Scheduling For Multi-Homed On-Board Networks", MOBIARCH'09 in ACM MOBISYS'09, Krakow, Poland, 22 June 2009.

NEMO (Network Mobility Standard from IETF) Downlink load balancing at Mobile Router Home Agent (HA)

Sudden change in bandwidth at the entry to a new location may overload a link causing buffer overflow at the cellular tower

Load Balancing Algorithm

  Balance load using Proportional Fair Scheduler (assign load to a link proportional to its bandwidth capacity)

  No geointelligence: estimate bandwidth every 2 sec and reshuffle loads if necessary

  With geointelligence: continue as before, but upon entering a new location, fetch bandwidth information from geointelligence and reshuffle load if necessary

Simulation Model

  Application   64Kbps Audio steaming (G771 Codec)   Poisson streaming session arrival   Exponential session duration (mean 2 minutes)

Measuring User Perceived QoS   We use Mean Opinion Score (MOS)

  Packet loss statistics (loss rate, burst size, etc.) are converted to MOS using ITU E-model

  If MOS drops below 3, we will call it a `glitch’ (because it will ‘annoy’ the user)

MOS Quality Impairment 5 Excellent Imperceptible 4 Good Perceptible 3 Fair Slightly Annoying 2 Poor Annoying 1 Bad Very Annoying

ITU E-Model

R factor:

Ro: Basic signal-to-noise ratio, Ro =93.2 for G711. Is: impairments which occur with the voice signal, set to 0. Id: impairments caused by delay and the effective equipment impairment factor, set to 0 A: Compensation of impairment factors, set to 0 Ie-eff: impairments due to packet-losses of random distribution.

Ppl: Packet-loss Probability Packet-loss Robustness Factor Bpl =25.1 for G.711.A (with PLC) BurstR: Average Burst Length of Burst Lost Packets

E-Model (cont.)

For R < 0:

For 0 < R < 100:

For R > 100:

E-Model (cont.)

Converting R factor to MOS:

Average Number of Glitches per Trip

100-300% more glitches if no geointelligence used

Conclusion (1)

  Location, even at 500m scale, seems to influence 3G bandwidth (bandwidth is geo-sensitive at many scales)

  Is bandwidth geo-sensitivity just a Sydney phenomenon?   No. See “performance comparison of 3G and metro-scale wifi for

vehicular network access”, ACM International Measurement Conference 2010 --- confirming geo-sensitivity for New York roads

  Using our ‘bandwidth entropy’ method, they observed geo-sensitivity even at 10m scale for 3G as well as WiFi

Conclusion (2)

  Even simple geointelligence of past average of a location seems to provide significant improvement for streaming quality of experience (at least for TFRC platform)

  Are these improvements only applicable to TFRC?   No. NOKIA has recently demonstrated that 3GPP streaming standard

can also benefit from such geointelligence. See “Geo-predictive real-time media delivery in mobile environment”, in ACM Mobile Video Delivery (in conjunction with ACM Multimedia 2010)

Future Directions

  How to gather geointelligence for every roads on this earth?   We are currently working on a croudsourcing concept

  What’s the best way to integrate geointelligence in streaming or other application platforms   Many issues to consider --- client driven or server driven, user’s

location privacy, impact on installed base, etc.

Key Publications http://www.cse.unsw.edu.au/~vnet

  J. Yao, S. Kanhere, and M. Hassan, ”Improving QoS in high-speed mobility using bandwidth maps", IEEE Transactions on Mobile Computing (in press)

  J. Yao, S. Kanhere, and M. Hassan, "Quality Improvement of Mobile Video Using Geo-intelligent Rate Adaptation", IEEE WCNC 2010, Sydney, 18 April 2010.

  J. Yao, S. S. Kanhere, M. Hassan, "Geo-intelligent Traffic Scheduling For Multi-Homed On-Board Networks", MOBIARCH'09 in ACM MOBISYS'09, Krakow, Poland, 22 June 2009.

  J. Yao, S. Kanhere and M. Hassan, "An Empirical Study of Bandwidth Predictability in Mobile Computing", WiNTECH’08 in ACM MOBICOM 2008, San Francisco, 19 September 2008.