Studying Internet routing data can turn up some interesting trends. Recently I have been looking into routing and hosting of sites under various country code Top-Level Domains (ccTLDs). The Internet Society examines the impact of local Internet Exchange Points (IXPs) in many contexts, which has engaged me to investigate the intersection of both.
When writing the report, I took a trans-disciplinary approach to the task as there are technical aspects to the data like latency, hop count, and the hosting of Autonomous System Numbers (ASNs). What also stood out for me was that these domains represented cultural elements that we in network engineering either ignore or avoid.
One important finding in the data is that business models and operating models matter at this level of analysis. From an engineering standpoint, an operating model is what the hosting company or network has done to build and deliver its services.
For some operators, the entirety of their model has them in one or two locations far from the economy in question. There is no benefit to end users in terms of low latency — it’s actually very high, or access over a local IXP, where there’s no benefit at all.
Meanwhile, the extreme opposite of that scenario was an operator with anycast DNS and proxy servers present across many of the IXPs. Latency was always low, and the metrics always were great for the end user. Also, two other groupings became evident in the analysis.
The big cloud offerings were ‘everything to everyone’ or ubiquitous, yet the metrics were widely varied. Their business model is not about selling optimal routing to their clients. If anything, those decisions are left to their client to decide. The result is that ccTLD content resides all over their networks across the region, and the world.
The last and most important grouping was the local organizations in economies like Japan, Malaysia, and Indonesia. This is where the cultural element becomes clearer. These websites are in the local language, the domains are registered in the local ccTLD, and the site is hosted within a local data centre.
That last scenario is also very good news for the local IXP. The latency metrics for Japan, Indonesia, and Malaysia were 6.17ms (91.34%), 10.78ms (89.62%), and 5.40ms (81.59%). Those figures would not be possible without good access to Internet peering.
When I looked at the Australian data, I was surprised to find Australian websites (per number of domains) in the majority were hosted overseas in Singapore, UK, USA, and Canada. The Ng and Taneja 2019 study found web-use-similarity clusters. In the five-cluster analysis, where Australians browse matches where hosts reside.
In Ng and Taneja’s 18-cluster analysis there are key alignments within South East Asia and South Asia, but notably, Singapore aligns with the same usage patterns as Western Europe and the USA. Australia’s 18-cluster group includes New Zealand, Canada, Portugal, and Scandinavia.
This brings up a more counter-intuitive concept. Is localization on the Internet a possible factor here? Well, yes, the data supports that very well. The Japan, Malaysia, and Indonesia metrics support the idea that the content is 80-90% local on several factors. Analysis of the ownership of the ASN brings up another factor — there are active local companies providing those hosting services.
Indonesia, for example, has 75 foreign ASNs hosting 34,986 domains versus over 600 local ASNs hosting 42,909 domains. The local providers have more market share collectively and they outnumber the multinationals and foreign operators eight to one. Localization as a concept is much more heavily nuanced for all ccTLDs.
So, what does this mean in the holistic sense for ccTLDs and IXPs? Local peering and local hosting are technical operations that rely on each other but also on local firms, local content creation, and local Internet users. Not all operating and business models suit this environment, some even work against the IXP dynamic.
Overall, what is really gluing this all together is the cultural linkage between the content and the consumer. In symbiosis between these two parties are local IXPs, local hosting infrastructure, and Internet Service Providers — firms with a business model aligned to the technical and cultural requirements and an active ccTLD.
My advice for governments and other development agencies in view of the above is simple: Start with the content and finish with a local IXP that has connected all of the local operators.
Watch Terry’s APRICOT 2022 presentation on this topic:
Terry Sweetser is a Senior Engineer with io Networks, has 30+ years of ITEE/Telco experience, and is well known in peering and governance forums.
The views expressed by the authors of this blog are their own and do not necessarily reflect the views of APNIC. Please note a Code of Conduct applies to this blog.