MOBILE DATA OFFLOADING : HOW MUCH CAN WIFI DELIVER?

Presented by Gregory Teodoro

Paper by Kyunghan Lee, Injong Rhee, Joohyn Lee, Song Chong, and Yung Yi

What is this about?• Mobile Data Offloading is the use of networking

technologies to deliver data originally mean for cellular networks.

• This paper is a quantitative study on performance of 3G mobile data offloading through WiFi Networks• Study done using over 100 iPhone users and data collection.

• Purpose is to discover how much of an effect data offloading has on mobile data traffic and battery life.

Introduction• Mobile data traffic is growing at unprecedented rates.

• Prediction is that by 2014 an average broadband mobile user will use 7GB of traffic each month, almost 5.4 times as much as is used now.

• Prediction that 66% of this data is through mobile video data.• Proposed solutions to this problem have issues.

• Scaling network capacity by building more towers and base stations, or upgrading stations comes at a huge cost, with no gain.• Revenue is independent from actual data usage.

• Switch to pure-usage pricing.• May backfire, as it singles out particular user groups.

WiFi Offloading instead?• Most viable solution at moment. Why?

• Building WiFi hot spots is significantly cheaper.• Can piggy back off a user’s own WiFi AP.• Already a wide-spread deployment of WiFi APs.• Addresses the “Time-to-Capacity” issue for current needs of

additional WiFi

Types of Offloading• On-The-Spot

• Uses spontaneous connectivity to WiFi and transfer data on the spot.

• When a user leaves WiFi coverage, offloading ends and unfinished transfers through cellular networks.

• Smart phones already give priority to WiFi than cellular interface.• Delayed

• Each data transfer is given a “deadline” when it must be sent out.• Sends the data piece by piece as a user enters and exits different

WiFi areas.• If data is not sent out before deadline, it is finished using the

cellular networks.

User and Network Payoffs• How does this help the Users and Network Providers?

• Using WiFi to offload data lowers overall cost of data transfers.• Users may benefit from lowered subscription prices due to lowered

costs.• Proper use of data transfer delays via delayed offloading can help

users select more specific plans.• Fundamentally tied to mobility patterns and WiFi availability.

Findings Summary

• On-the-spot offloading can offload ~65% of total traffic load.• This is without using delayed offloading at all.

• Delayed Offloading only gained 2-5% efficiency when 100 seconds used• Admittedly incredibly different from other findings on the same idea.

• When upwards of an hour is used, the gain becomes ~29%.• On-the-spot offloading can achieve 55% energy savings due to

reduction in transfer times.• Once again, 100 second delays offer only 3% energy savings gain.• Increasing delays to an hour the gain increases by 20%.

• A prediction based offloading strategy (such as Breadcrumbs) must predict over several minutes to be useful.• Interconnection time can be as long as 40 minutes, making prediction

hard.• Average completion time of data transfers is much shorter than

delay deadlines.• Even with delayed offloading, uploading a 30 MB video is still faster than

using a 3G network.

Experiment Setup• Uses an application called Dtap that records WiFi

connectivity, and sends recorded data to servers.• Scanned every 3 minutes for an AP.• Records GPS location as well as duration, data rate, and time.• Does not perform offloading in and of itself.

• Why? Offloading for arbitrary data such as this drained too much battery.

• 97 volunteers who own iPhone 3G/3GS in Korea were asked to use Dtap for a period of 18 days.• Diverse occupational background and various major cities used.

• Collected 705 valid daily traces.

Temporal coverage per user, time and hourly mobility.

Key Observations• Temporal Coverage

• Performance of offloading highly depends on the time duration a user stays in a WiFi covered area.

• Average coverage across all users are 70% for all day, and 63% for daytime only.• Differences caused by users who are more likely to have WiFi coverage

at home.• Different from other findings. Why?

• Other reports use measurements through war-driving, and do not account for natural mobility of a user.

• Users typically spend far more time at home or an office than traveling.• To prove this, they record traveling distances as well. (See Figure 3c)

Key Observations (Cont.)• Spatial coverage measured as well.

• The fraction of an area that is under any WiFi coverage.• Only given a rough estimate as users do not naturally walk around

the entire area.• Does give a usable lower bound.

• Conclusion : About 8.3% is spatially covered.• Combined Findings

• Temporal coverage is 3.5~8 times larger than spatial coverage. What does this mean?• Indicates that most users stay inside a WiFi network for a long time

once connected.• Average connection time is 2 hours for all day, and 52 minutes for daytime

only.

•

Findings (Cont.)• End-to-End Rates

• Average data rate is ~1.97Mbps• Average is however highly skewed by night time.

• Highest data rates are during night, around 2.76 Mbps, and 1.26 Mbps during the day.• Why? Users are more likely to be connected to personal, home APs

during the night.• Findings point to offloading during the night to be very effective

because of this.• Provided users can accept a delay of data transfer until night.

• User mobility has a lower correlation with data rate than temporal coverage.

Offloading Efficiency• With the findings and tracings, a simulation can be

created.• Using the simulation, Offloading Efficiency is defined as the total

bytes transferred via WiFi, divided by all the bytes generated.• To further understand mobile traffic, projection data

released from CISCO is used.

Offloading Efficiency (Cont.)• The amount of traffic offloaded to WiFi from a 3G network is

then measured.• Experiment assumes that all transfers of video and data are delayed.

• Surprisingly, on-the-spot offloading also achieves extremely high offloading efficiency.

• Further more, due to the above, if most mobile traffic is placed onto smart phones, 65% of data traffic can be offloading to WiFi automatically.

• Why? Average users spend more time in WiFi zones than traveling between them ala war-driving.

• With delayed added, offloading efficiency increases substantially.• 100 seconds or less was negligible.• Long deadlines can bring efficiency up to 88%

• Admittedly unrealistic, as users may ignore delayed transfers and opt for on-the-spot only.

Completion Time• Deadlines of 30 minutes to an hour may be unrealistic.

• Findings do indicate that most transfers finish long before this.• Example. Photos given a 60 second deadline finished only 6 seconds

after the same transfer without offloading.• For larger files, users may complete in the same time with delayed

offloading as they would using no offloading at all.• Delayed Offloading has a longer completion time than on-the-spot

offloading.• However it uses 3G networks far less.• Under certain circumstances (bad or little WiFi access), delayed can be

faster than on-the-spot.

Energy Savings• There is a fundamental trade-off between energy

consumption and delay transfers.• 3G networks more widely available, but transfer slower.

• Result? More battery powered is used to transfer a file.• Using delay offloading, WiFi can be used, so less time and battery

is spent on transferring.• Energy consumption per minute using 3G or WiFi is

roughly the same• Main difference comes in as a difference in transfer time.

Findings

Traffic Types• Bursty, small-sized traffic benefits the least from delayed

offloading.• Larger, bigger-sized traffic got the most benefit from

delayed offloading.• In terms of type of traffic, texts and photos benefited the

least from delayed, and more from on-the-spot.• Video and Multimedia backups benefited the most from

delayed offloading.• Regardless of data type, both benefited from some form

of offloading.

Impacts of WiFi Deployment• For simulation, the deployment found during testing was

used, then slowly thinned out by eliminating APs.• Two elimination methods were used, one random, and

one based on the activity of an AP.• Eliminating half of the APs using the one based on activity, getting

rid of the last used first had little effect.• Eliminating half of the APs through a random means halved

offloading efficiency. Why?• Most of the traffic went through popular APs, such as coffee shops,

offices, or other public APs.

• Findings. Implies that careful deployment plans can yield improvements in capacity without increasing density.

On this paper…• Strengths

• Relatively large sized study, and used that information to run effective simulations.

• Very detailed reports on their findings• Many different visual means to convey information was used.• Informative and keeps a constant comparison to other papers in the same field.

• Weaknesses• Rather redundant, some points are mentioned and proven multiple times.• References other papers via name/number, but does not offer a quick blurb as

to what the other paper had found.• Are the findings useful? Paper admits offloading is already handled by smart

phones automatically.• Mentions that findings give concrete evidence that mobile companies can save money

on data transfers and how much they approximately can save, but why would a company lower prices anyhow?

• Does not really consider efficiency in areas that are not urban, or don’t have easily accessible public WiFi APs.

Future Work• Paper was relatively new, coming out in 2010.• Events since its release.

• KDDI (Japan) rolled out the world’s largest Wi-Fi based offload network.• http://www.mobiledataoffloading.com/mobile-data-offload-news/520/

• First Wi-Fi system that can deliver Gigabit capacity and outdoor speeds of 450 MBPS released by Winncom.• http://www.mobiledataoffloading.com/mobile-data-offload-news/504/

• Company iPass formed, offering Open mobile Exchange Platforms to integrate Wi-Fi with 3G and 4G Networks.• http://www.mobiledataoffloading.com/mobile-data-offload-news/498/

• Next Offloading Conference will be in the 26th of March 2012.