Visualizing the Digital Divide from an Internet Point of View & Challenges Prepared by: Les Cottrell SLAC Umar Kalim NIIT, Shahryar Khan NIIT, Akbar Mehdi NIIT For COMSATS University, Islamabad, Pakistan, March 14,

Outline Digital Divide: –Examples of effect of Digital Divide & why it matters –How we measure it –What we find A network challenge for mathematicians, statisticians

Why Does it Matter 4. Sep 05, international fibre to Pakistan fails for 12 days, satellite backup can only handle 25% traffic, call centres given priority. Research & Education sites cut off from Internet for 12 days Heloise Emdon, Acacia Southern Africa UNDP Global Meeting for ICT for Development, Ottawa July 3. Primary health care giver, somewhere in Africa, with sonar machine, digital camera and arrangement with national academic hospital and/or international health institute to assist in diagnostics. After 10 dial-up attempts, she abandons attempts to connect 1.School in a secondary town in an East Coast country with networked computer lab spends 2/3rds of its annual budget to pay for the dial- up connection. –Disconnects 2. Telecentre in a country with fairly good connectivity has no connectivity –The telecentre resorts to generating revenue from photocopies, PC training, CD Roms for content.

How do we measure it? PingER project Arguably the world’s most extensive active end-to- end Internet Performance Project

PingER Methodology Internet 10 ping request packets each 30 mins Remote Host (typically a server) Monitoring host > ping remhost Ping response packets Measure Round Trip Time & Loss Data SLAC Once a Day Uses ubiquitous ping

Architecture Monitor hosts send 21 pings each 30 mins to Remote Hosts and cache results Archive hosts gather data daily, save, analyze & make results available publicly via web

PingER Deployment PingER project originally (1995) to measure network performance for US, Europe and Japanese HEP community Extended this century to measure Digital Divide: –Collaboration with ICTP Science Dissemination Unit –ICFA/SCIC: Monitor 44 sites in S. Asia >120 countries (99% world’s connected population) >30 monitor sites in 14 countries

World Measurements: Min RTT from US Maps show increased coverage Min RTT indicates best possible, i.e. no queuing >600ms probably geo-stationary satellite Between developed regions min-RTT dominated by distance –Little improvement possible Only a few places still using satellite for international access, mainly Africa & Central Asia

Effect of Losses Losses critical, cause multi-second timeouts Typically depend on a bad link, so ~distance independent > 4-6% video-conf irritating, non-native language speakers unable to communicate > 4-5% irritating for interactive telnet, X windows >2.5% VoIP annoying every 30 seconds or so Burst losses of > 1% slightly annoying for VoIP

Losses from SLAC to world >=12% >=5% <12% >=2.5% < 5% >=1% < 2.5% < 1% # hosts monitored increased seven-fold Increase in fraction with good loss –Despite adding more hosts in developing world

Loss Improvement by Population Loss by country weighted by population of country

Unreachability All pings of a set fail ≡ unreachable Shows fragility, ~ distance independent Developed regions US, Canada, Europe, Oceania, E Asia lead –Factor of 10 improvement in 8 years Africa, S. Asia followed by M East & L. America worst off Africa NOT improving US & Canada Europe E Asia C Asia SE Europe SE Asia S Asia Oceania Africa L AmericaM East Russia Developed Regions Developing Regions

World thruput seen from US Behind Europe 6 Yrs: Russia, Latin America 7 Yrs: Mid-East, SE Asia 10 Yrs: South Asia 11 Yrs: Cent. Asia 12 Yrs: Africa South Asia, Central Asia, and Africa are in Danger of Falling Even Farther Behind Throughput ~ 1460Bytes / (RTT*sqrt(loss)) (Mathis et al)

Normalized for Details Note step changes Africa v. poor S. Asia improving N. America, Europe, E Asia, Oceania lead

Overall (Aug 06) ~ Sorted by Average throughput Within region performance better (black ellipses) Europe, N. America, E. Asia generally good M. East, Oceania, S.E. Asia, L. America acceptable C. Asia, S. Asia poor, Africa bad (>100 times worse) Monitored Country

South Asia Population

S Asia Bandwidth & Internet use Note Log scale for BW India region leader Pakistan leads bw/pop Nepal very poor Pakistan leads % users Sri Lanka leads hosts% Pakistan leads bw/pop Nepal, Bangladesh, Afghanistan very poor

S Asia PingER Coverage Monitor 44 sites in region. 6 Monitoring hosts (3 ea in India & Pakistan) Loss from CERN Min-RTT from CERN

Divides into 2 –India, Maldives, Pakistan, Sri Lanka –Bangladesh, Nepal, Bhutan, Afghanistan Weekend vs. weekday indicates heavy congestion Derived thruput

Digital Access Index (DAI): Infrastructure availability, Affordability of access, Education, Quality of ICT, & Internet usage Europe, E Asia (except China), Oceania top right Israel & Singapore with top group Middle East in middle, Iran poorest Africa bottom left S. Asia split: Bhutan, Nepal, Bangladesh with Africa India, Pak, Sri Lanka better Strong positive linear correlation, C Asia

DAI & S. Asia

D.D. Conclusions Last mile problems, and network fragility Decreasing use of satellites, expensive, but still needed for many remote countries in Africa and C. Asia Africa ~ 10 years behind and falling further behind, leads to “information famine” E. Africa factor of 100 behind Europe –EASSy project will bring fibre to E. Africa, hopefully better access than SAT3 Africa big target of opportunity –Growth in # users %, Africa 625% –Need more competitive pricing Fibre competition, government divest for access, low cost VSAT licenses Consortiums to aggregate & get better pricing ($/BW reduces with BW) –Need better routing - IXPs –Need training & skills for optimal bandwidth management Internet performance correlates strongly with UNDP & ITU development indices –Increase coverage of monitoring to understand Internet performance

Challenge, however… Elegant graphics are great to understand problems BUT: –Can be thousands of graphs to look at (many site pairs, many devices, many metrics) –Need automated problem recognition AND diagnosis So developing tools to reliably detect significant, persistent changes in performance –Initially using simple plateau algorithm to detect step changes

Some are seasonal Others are not Events may affect multiple-metrics Misconfigured windows New path Very noisy Examples of real data Seasonal effects –Daily & weekly Caltech : thrulay Nov05 Mar Mbps UToronto: miperf Nov05 Jan Mbps UTDallas Pathchirp thrulay Mar Mar iperf Mbps Events can be caused by host or site congestion Few route changes result in bandwidth changes (~20%) Many significant events are not associated with route changes (~50%)

Changes in network topology (BGP) can result in dramatic changes in performance Snapshot of traceroute summary table Samples of traceroute trees generated from the table ABwE measurement one/minute for 24 hours Thurs Oct 9 9:00am to Fri Oct 10 9:01am Drop in performance (From original path: SLAC-CENIC-Caltech to SLAC-Esnet-LosNettos (100Mbps) -Caltech ) Back to original path Changes detected by IEPM-Iperf and AbWE Esnet-LosNettos segment in the path (100 Mbits/s) Hour Remote host Dynamic BW capacity (DBC) Cross-traffic (XT) Available BW = (DBC-XT) Mbits/s Notes: 1. Caltech misrouted via Los-Nettos 100Mbps commercial net 14:00-17:00 2. ESnet/GEANT working on routes from 2:00 to 14:00 3. A previous occurrence went un-noticed for 2 months 4. Next step is to auto detect and notify Los-Nettos (100Mbps)

On the other hand Route changes may affect the RTT (in yellow) Yet have no noticeable effect on on available bandwidth or throughput Route changes Available Bandwidth Achievable Throughput

Seasonal Effects on events Change in bandwidth (drops) between 19:00 & 22:00 Pacific Time (7:00-10:00am PK time) Causes more anomalous events around this time

Forecasting Over-provisioned paths should have pretty flat time series –Short/local term smoothing –Long term linear trends –Seasonal smoothing But seasonal trends (diurnal, weekly need to be accounted for) on about 10% of our paths Use Holt-Winters triple exponential weighted moving averages

Econometrics Econometrists use forecasting techniques for predicting the behavior of economic metrics –Auto Regressive Integrated Moving Average (ARIMA & ARMA) –Very mathematical, multiple techniques: Integration to make stationary Auto-regression Moving Averages Determining parameters etc. can be an art –Our (Fareena Saqib) first look at was promising Have a long document of how far we got –Do not currently have someone working on next steps.

Experimental Alerting Have false positives down to reasonable level (few per week), so sending alerts to developers Saved in database Links to traceroutes, event analysis, time-series

More information/Questions Acknowledgements: –Harvey Newman and ICFA/SCIC for a raison d’etre, ICTP for contacts and education on Africa, Mike Jensen for Africa information, NIIT/Pakistan for developing valuable tools, Maxim Grigoriev (FNAL), Warren Matthews (GATech) for ongoing code development for PingER, USAID MoST/Pakistan for development funding, SLAC for support for ongoing management/operations support of PingER PingER –www-iepm.slac.stanford.edu/pinger, sdu.ictp.it/pinger/africa.htmlwww-iepm.slac.stanford.edu/pinger sdu.ictp.it/pinger/africa.html Human Development – Case Studies: – Sahara+Case+Studyhttps://confluence.slac.stanford.edu/display/IEPM/Sub- Sahara+Case+Study –

Extra Slides Follow

Costs compared to West Sites in many countries have bandwidth< US residence –“10 Meg is Here”, Africa: $5460/Mbps/m –W Africa $8K/Mbps/m –N Africa $520/Mbps/m Often cross-country cost dominates cf. international 1 yr of Internet access > average annual income of most Africans, Survey by Paul Budde Communnications

UNDP Human Development Index (HDI) A long and healthy life, as measured by life expectancy at birth Knowledge, as measured by the adult literacy rate (with two-thirds weight) and the combined primary, secondary and tertiary gross enrolment ratio (with one-third weight) A decent standard of living, as measured by GDP per capita. Africa PingER - Strong Correlation - Non subjective - Quicker / easier to update

Med. & Africa vs HDI N. Africa has 10 times poorer performance than Europe Croatia has 13 times better performance than Albania Israel has 8 times better performance than rest of M East Med. Countries E. Africa poor, limited by satellite access W. Africa big differences, some (Senegal) can afford SAT3 fibre others use satellite Great diversity between & within regions

Why does it matter: Business G8 specifically pledged support for African higher education and research by “Helping develop skilled professionals for Africa's private and public sectors, through supporting networks of excellence between African's and other countries' institutions of higher education and centres of excellence in science and technology institutions” Saturating western markets High growth IT markets: BRIC NOT business as usual –New business models –Distinct needs –Dearth of distribution channels Traditional MNC Business Model >$20K per year 75 to 100 million people Some MNCs >$1, K per year 1.5 to 1.75 billion people Local Firms <$1,500 per year 4 billion people Future Opportunity? Prahalad and Hart Karen Coppock RDVP, Stanford