IXPs and the Impact on International Transit in Africa
By Michuki Mwangi (ISOC)ITU Regional Economic and Financial Forum of Telecommunications/ICTs for Africa18 – 19 January 2016Abidjan, Côte d’Ivoire
What is the Internet Society?The Internet Society (ISOC) is a cause-based organization that works with governments, industries, and others to ensure the technologies and policies that helped develop and evolve the Internet will continue into the future.
Our programs cultivate an Internet that is open to everyone, everywhere and aim to ensure that it will continue to be a tool for creativity, innovation, and economic growth.
MISSION: To promote the open development, evolution, and use of the Internet for the benefit of all people throughout the world.
VISION: The Internet is for everyone
Global Presence
100+Chapters Worldwide
65,000+Members andSupporters
145+Organization Members
6Regional Bureaus
18 Countries with ISOC Offices
NORTH AMERICA
LATIN AMERICA & CARIBBEAN
EUROPE
AFRICA
THE MIDDLE EAST
ASIA
OCTOBER 2013Chapters
Agenda
Background on Africa Interconnection
Impact of Peering on International Transit
Conclusion
Background on Africa Interconnection:
2008 Fiber infrastructureAgainst IXPs in Africa
• 17 IXPs in 15 African Countries• 12 were considered responsive based
on a survey conducted in 2008• Average number of years across all
IXPs was 4.1yrs• 10 of 12 IXPs provided traffic stats
• Highest had 200Mbps and lowest was 300Kbps
• 1/3 of the IXPs had an open membership policy. Remaining 2/3’s were subject to regulatory or membership set criteria
• 75% of the respondents had a Mandatory Multilateral peering policy
• Copper and wireless connectivity was predominant over fiber
• 41% of the IXPs did not charge fees and the highest fees were $9,000 and lowest was $50
• 75% were ran by ISPA. Others by NREN, Govt. and by a not-for-profit (non-ISPA)
1993 – 2001: SAT2 Total Capacity 560Mbps
2001 – 2008: SAT3 Total Capacity 340Gbps
2001 - 2008: SAFE Total Capacity 440Gbps
2000 – 2005: SEA-ME-WE-3 Total Capacity 480Gbps
2008: Traffic distribution across10 African IXPs
What we observed from 2008 Survey
Positive
• There was a seemingly good linkage between IXPs and the ccTLD
• IXPs had relatively “good” connectivity
• The role of Governments in IXPs was seemingly progressive
• IXP Models were ideal “at that stage”
Negative
• The Mandatory Multi-Lateral peering policy was a dominant obstacle to IXP growth
• Membership policies were not progressive
• IXP sustainability was an area of concern.
• More value added services need to be implemented at the IXPs
Gaps identified
• There was a lack of general knowledge on IXP best practices to enable them grow
• There was no platform for sharing experiences or learning amongst the IXPs and operators
• Cross-border interconnection between networks was almost non-existent
• 30% of the known IXPs were unresponsive for unknown reasons
• Central, North and West Africa were lagging behind in IXP deployment
• Africa has a large “Internet Transit Deficit” in comparison to other regions.
Impact on Africa’s Interconnection Scene
• Conducted 15 workshops in 15 Africa countries on Technical and policy issues related to IXP development
• Workshops trained over 300 engineers and policy makers
• Established and organized 6 AfPIF events
• Supported the establishment and activities of the Africa IXP association (Af-IX)
• Provided technical assistance, equipment support to at least 10 IXPs
• Created awareness with key policy stakeholders i.e UNECA, African Union, Regional Regulatory Associations, Regional Economic Communities (RECs)
• Overall growth of the African peering ecosystem
• Work lead to award to implement African Union AXIS project
AXIS Project
• African Union awarded Internet Society two (2) contracts to implement AXIS project between 2012 – 2014
• Project focused on Capacity building and technical assistance in establishing IXPs
• Overall, phase 1 of the AXIS project trained over 1,200 people from 28 African Countries over the 2 year period
• AXIS Best practice workshops covered 28 African countries and trained over 700 people in English, French and Portuguese
• AXIS Technical Aspects workshops covered 28 African countries and trained over 500 engineers in interconnection technologies in English, French and Portuguese
• Phase 2 of AXIS conducted 5 workshops across the 5 geographic regions of Africa (Central, East, North, South and West Africa)
• Engaged over 350 experts from relevant public and private sector to discuss regional interconnection issues in in English and French
AXIS Outcomes
• Phase 1 Outcomes• 8 new IXPs have been established as a result of the AXIS project• Most successful IXP project in the region and raised importance of IXPs
across the board• Established new partnerships to enable deliver workshops in different
languages (French & Portuguese) • In the process, we also developed regional francophone IXP expertise
with support of partners
• Phase 2 Outcomes• At least 2 IXPs per region will receive financial assistance to enable
them evolve and grow to become regional IXPs• Project will provide technical assistance to ISPs to enable them evolve
and become regional carriers• Project established and supported work of 5 regional task-forces to
review policy and regulatory barriers that inhibit national and cross-border interconnection in their respective regions.
Cross-Border: Percentage of ASNs (origin ASNs) by country assignment in routes collected
All ASNs Origin ASNs
IXP: JINX (ZA)
Source: Roderick Fanou Presentation at AfPIF-2015
Cross-border: Percentage of ASNs (origin ASNs) by country assignment in routes collected
All ASNs Origin ASNs
IXP: CAIX (EG)
Source: Roderick Fanou Presentation at AfPIF-2015
Cross-border: Percentage of ASNs (origin ASNs) by country assignment in routes collected
All ASNs Origin ASNs
IXP: MIX (MZ)
Source: Roderick Fanou Presentation at AfPIF-2015
Example: ASN’s visible at KIXP
All ASNs Origin ASNs
IXP: KIXP (KE)
Source: Roderick Fanou Presentation AfPIF-2015
• 34 IXPs in 26 Countries or 42% increase since 2008
• Total Traffic exchanged at African IXPs >160Gbps
• More IXPs in West & Central Africa• Data shows more cross-border
interconnection/peering in East and Southern Africa
The Impact of Peering on International Transit
Reference data
• The global average connection speed is 5Mbps while the global average peak connection speed is 29.1 Mbps.
• Africa’s International Internet bandwidth has been growing at >50 CAGR 2011 - 2015
• International transit prices have been declining by >25% CAGR 2010 – 2015 in most major cities
• Internet traffic volume driven by video grows at a rate of ~60% every year
• Watching a HD stream requires ~6Mbps connection while watching an SD stream requires ~2Mbps connection
• Watching movies or TV shows on Netflix uses about 1 GB of data per hour for each stream of standard definition video, and up to 3 GB per hour for each stream of HD video
• Africa imports more than >95% of its Internet traffic from outside the continent. Most of the content is imported from Europe
Scenario Assumptions
• Bandwidth costs excluding backhaul, cross-connects and other expenses
• The service providers are offering 1 Mbps service to home users at a contention ratio of 50:1
• The local capacity/circuits cost less than international link
• Dedicated International capacity links pricing starting at STM 1 capacity starts $120 /Mbps
• Local point-to-point links pricing start at STM1 capacity starts at $30/Mbps
• 10% of the heavy users consume ~60% of the bandwidth
Assumption 1: International Transit Costs
ISP TransitCapacity
$ Transit Price / Mbps
$ MonthlyTransit Cost
Potential Customers at 50:1 contention ratio
Contended Transit Price / Mbps
ISP A 155 (STM1) $ 120 18,600 7,750 $ 2.40
ISP B 1GE $ 100 100,000 50,000 $ 2.00
ISPC 10GE $ 80 800,000 500,000 $ 1.60
Assumption 2: Local Peering Costs
ISP Peering Capacity
Peering Price /Mbps
$ Monthly Peering cost
Potential Customers at 50:1 contention ratio
Contended PeeringPrice/Mbps
ISP A 155 $ 30 4,650 7,750 $ 0.60
ISP B 1GE $ 20 20,000 50,000 $ 0.40
ISP C 10GE $ 10 100,000 500,000 $ 0.20
Transit vs. Peering Ratios – Scenario 1: 95% Transit 5% Peering
ISP TransitBandwidth Savings (Mbps)
New Potential customers at 50:1 (transit) contention
Total NewCustomers
Contended Aggregate Price / Mbps
Monthly Income
Monthly Expense (Transit + Peering)
Monthly Cost Savings
ISP A 7.75 387.5 8137.5 $ 3.0 $ 24,412.5 $ 23,250 $ 1,165.2
ISP B 50 2,500 52,500 $ 2.40 $ 126,000 $ 120,000 $ 6,000
ISP C 500 25,000 525,000 $ 1.80 $ 945,000 $ 900,000 $ 45,000
Transit vs. Peering Ratios – Scenario 1: 95% Transit 5% Peering - Observations
• With 5% of traffic going through peering there would be no contention experienced for local traffic (when peering links equal transit links)
• However, for all ISPs to increase their monthly profit margins, they would need to reduce the capacity of their peering links to at most 5% of their Transit links
• Reducing the peering link capacity to 5% will reduce the local traffic contention ratio to less than 50:1
• All ISPs are able to realize modest cost savings resulting from additional customers but no Transit upgrade costs
Transit vs. Peering Ratios – Scenario 2: 80% Transit 20% Peering
ISP TransitBandwidth Savings (Mbps)
New Potential customers at 50:1 (transit) contention
Total NewCustomers
Contended Aggregate Price /Mbps
Monthly Income
Monthly Expense (Transit + Peering)
Monthly CostSavings
ISP A 31 1,550 9,300 $ 3.0 $ 27,900 $ 23,250 $ 4,650
ISP B 200 10,000 60,000 $ 2.40 $ 144,000 $ 120,000 $ 24,000
ISP C 2,000 100,000 600,000 $ 1.80 $ 1,080,000 $ 900,000 $ 180,000
Transit vs. Peering Ratios – Scenario 2: 80% Transit 20% Peering - Observations
• With 20% of traffic going through peering there would be no contention experienced for local traffic (when peering links equal transit links)
• However, for all ISPs to increase their monthly profit margins, they would need to reduce the capacity of their peering links to at most 20% of their Transit links
• Reducing the peering link capacity to 20% will reduce the local traffic contention ratio to less than 50:1
• All ISPs are able to realize better cost savings resulting from more new customers without transit upgrade costs
• ISPs/Operators that bill on usage, are likely to realise a 20% increase in billed revenue on
Transit vs. Peering Ratios – Scenario 3: 50% Transit 50% Peering
ISP TransitBandwidth Savings (Mbps)
New Potential customers at 50:1 (transit + peering) contention
Total NewCustomers
Contended Aggregate Price /Mbps
Monthly Income
Monthly Expense (Transit + Peering)
MonthlyCost Savings
ISP A 77.5 3875 11,625 $ 3.0 $ 34,875 $ 23,250 $ 11,625
ISP B 500 25,000 75,000 $ 2.40 $ 180,000 $ 120,000 $ 60,000
ISP C 5,000 250,000 750,000 $ 1.80 $ 1,350,000 $ 900,000 $ 450,000
Transit vs. Peering Ratios – Scenario 3: 50% Transit 50% Peering - Observations
• With 50% of traffic going through peering there would be no contention experienced for local traffic (when peering links equal transit links)
• ISPs/Operators that bill on usage are likely to realize significant increased revenues compared to those who do not .
• However, for all ISPs to increase their monthly profit margins, they would need to reduce their peering links capacity to at most 50% of their Transit links
• Reducing the peering link capacity to 50% will subject the local traffic contention ratio of 50:1
• High contention ratio will negatively impact the end user experience on accessing local traffic.
Transit vs. Peering Ratios – Scenario 4: 20% Transit 80% Peering
ISP TransitBandwidth Savings (Mbps)
New Potential customers at 50:1 (transit & 45:1 peering) contention
Total NewCustomers
Contended Aggregate Price /Mbps
Monthly Income
Monthly Expense (Transit + Peering)
MonthlyCost Savings
ISP A 124 6,200 13,950 $ 3.0 $ 41,850 $ 27,850 $ 14,625
ISP B 800 40,000 90,000 $ 2.40 $ 216,000 $ 140,000 $ 76,000
ISP C 8,000 400,000 900,000 $ 1.80 $ 1,620,000 $ 1,000,000 $ 620,000
Vision
2020
Transit vs. Peering Ratios – Scenario 4: 20% Transit 80% Peering - Observations
• With 80% of traffic going through peering there be a need to upgrade the local peering links to be twice their transit links
• doubling local peering link capacity will reduce the contention ratios to about 45% and improve end user experiences for local traffic
• With more traffic being local and maintaining the transit link capacity, the contention ratio on international traffic will be non-existent
• With more local traffic and increased revenues, it is potentially easier to reduce contention ratios as low as 10:1 and maintain positive bottom lines
• ISPs/Operators that bill on usage are likely to realize significant increased revenues compared to others
Conclusion
• Peering enables ISPs to saves costs from delays in upgrade of International transit capacity, despite addition of new users on the network.
• The transit vs. peering ratio impacts the ISPs bottoms line. An ISPs that peer more realize higher cost savings.
• A network can have fewer users with fast Internet access and require the same transit/peering capacity as a network with more users on slow Internet access
• High profit margins favor ISPs that have large bandwidth (Transit+ peering) capacity
• To realize significant profit margins, ISPs need to peer at least 50% of their transit capacity
• Transit usage/demand will continue to grow even after the ratios change in favor of peering traffic.
Thank [email protected]
http://www.internetsociety.org