Interconnection: Costs, Complexity, Policies Gaurab Raj Upadhaya Why Peering in Asia Sucks !

Disclaimer Everything in these slides is my own opinion.

Interconnection Markets Internet interconnections are concentrated in few hubs around the world – US West Coast (Silicon Valley, Los Angeles) – US East coast (New York, Ashburn) – London, Amsterdam, Frankfurt, Stockholm – Singapore, Hong Kong, Tokyo

Interconnection Markets It takes time and lots of efforts to create a viable interconnect market. – Singapore is a very recent entrant into this ‘hub’ status. It doesn’t always work at scale for everyone – Local relevance may sometimes be more value Moscow, Miami, Prague, Budapest Open and easy access are the key drivers – Why are Rio / Mumbai / Dubai not real hubs yet?

Let’s Dig Deeper into the Asian Hubs Singapore – A hub for a lot of other things – transport, banking, logistics due to ease of doing business – Equinix SG1 IX (287 networks according to peeringdb) Hong Kong – Similar to Singapore, traditional hub for logistics – Voice Interconnect hub, so carrier density – HKIX (297 networks according to peeringdb) Tokyo – The traditional jumping point to the ‘Internet’ in the US – Japan is one of the largest Internet markets (users, content, everything) – JPIX (184 ), JPNAP (150)

Greg’s Cable Map :

But Layer 1 infrastructure doesn’t translate into Layer 3 Interconnect Because of Cost Complexity Policies

Cost of Backhaul – NA/Europe Intra Europe 10G terrestrial wave Costs – USD for 10G between major points – USD for extended locations – Bucharest to London is ~2500KMs, costs below 2500 USD, or roughly /mbps-km. Intra-US 10G wave costs are similar (actually a bit higher) – US distances are longer – Large expanse of empty lands

Cost of Backhaul in APAC Intra Asia is mostly Sub sea, costs are traditionally higher, but going down drastically – Intra Asian Hub prices range from 12-20K/Month, depending on locations – Cost of local loop are a key driver, but with POP to POP design changes, the local loop is not a big consideration for many providers. Where it does, the cost of CLS to POP backhaul is a big problem

Cost of terrestrial Backhaul – North Asia Japan: Tokyo – Osaka : 10G ~-5-8K/month Korea: Seoul – Busan : 10G ~ 12K/month – South Asia India : Delhi-Mumbai: 10G ~50-60K a month Bangladesh : Dhaka-Cox’ 10G – 25K/month Nepal : Kathmandu-Pokhara 10G: - 10K/month – South East Asia Malaysia: Kualalumpur – Johor: 30K/month Thailand: Bangkok – Phuket: 10G ~25-30K/month *Philippines, Indonesia are mostly subsea. Different costs

Asian Transport Pricing 10 Gbps Wavelength Price Multiple over London-New York

Let’s look deeper CLS to POP costs – Tuas/ Changi – SG1 : 5-8K/month – Chikura/Maryuama - Otemachi : 5-8K/month – Hongkong : 4K/month premium for anywhere outside Mega-I. – Korea: it cost more for local backhaul from the CLS to POP, then the wet segment from Tokyo – India : Regulated Access charge ($4/mbps ?)

So, It is EXPENSIVE to get to peering locations – Transit may be cheaper – Going to LA is likely cheaper Operators keep costs artificially high – In Metro Areas – Offer IP Transit over the same transport, which is cheaper (vs. buying the transport). Complexity keeps the costs high – SDH – Weak Fiber Infrastructure (i.e, needing 3 routes to build full redundancy).

Complexity & costs are related In Tokyo metro area, metro 10G between certain Datacenters are available as low as $350. – Same transport from other providers range upto 5K/month Aggregation for other services are not optimal for data. – One major Indian mobile carrier aggregates into 3 locations, and those are not close to data gateways. – Network designed based around regulatory boundaries, and voice optimization Just easier to buy transit in 3 locations then try to aggregate and build peering. It adds the costs, makes engineering changes difficult.

It’s not just the cost or complexity Most Markets are closed except for HK – HKIX had an early start, and is popular even with the high local transport cost (imagine HKIX without transport cost.. without M-I).. – But then you have Dark Fiber in HK being limited to N- Waves Domestic Peering fabric are closed user groups or non existent – Korea, Japan, Vietnam, India have closed environment – In most other it’s likely non-existent

Can it be improved Singapore has nearly done it, so yes.. Can be done Things may move faster in JP, HK, SG then in other locations. But something more disruptive needed – Dark Fiber availability or metro waves availability – Metro area IXPs ? (in SG and HK)

Acknowledgement Telegeography Donald Clark Raphael Ho