====================================================================== Title: one or many internets? Date: 2024-11-16 Tags: box3, internet, philosophy Link: https://spool-five.com/box3/20241116t010854--one-or-many-internets__box3_internet_philosophy/ Word Count: 692 ====================================================================== #internet[1] =>[1] https://spool-five.com/box3/20251019t150941--internet__box3/ I was just reading a great post by Geoff Huston[2] about the IPv6 transition. =>[2] https://www.potaroo.net/ispcol/2024-10/ipv6-transition.html In it he discusses the idea that the original 1980's model of an internet with an end to end address space has been eclipsed by NAT techniques, CDNs, the increased importance of names over numbers, and so on. Image: https://spool-five.com/noteimg/ipv6-fig8.png The image above is taken from Geoff's post. It shows the recent decline in price for IPv4 addresses which, if you're thinking about internet architecture from a 1980's mindset, seems counter-intuitive. If more and more devices are being added to the network, devices which would traditionally have required their own unique IP address, then surely demand for IPv4 addresses should be _increasing_ (as it had been)? The truth is that how we use internet addresses has changed so much in recent years. There are estimated to be around 20 billion devices connected to the internet, but there are only around 3 billion usable IPv4 addresses. The old notion as the internet as a single, flat network where every device can directly connect to one another using a unique number is gone. This still works in theory (at the logical level) but in practice a lot of techniques are employed to stretch out those 3 billion addresses. The decline in value of IPv4 address could be a sign that these techniques are becoming the new norm. If we think of that original vision of the Internet as a flat, homogeneous network of nodes that can all connect to each other through some simple, 'dumb' behaviour and cooperation, then the modern internet has introduced what (at first) seems like a 3rd dimension, a level of 'depth'. Do we now have something like a 'tree' as opposed to a flat network, where NATs branch off into their own networks (and those branch off again, and so on)? In the philosopher Gilles Deleuze's terms, have we moved from a flat, rhizome-like structure to a hierarchical, tree-like structure? An observation might be that having 'multiple' networks, with various gateways between them would introduce performance costs. However, as the article above says, even that view is becoming a bit redundant because of CDNs and caching. Yes, there may have been a performance hit if you wanted to traverse across the whole network and multiple address translations, compared to just routing directly to an address, but that's not how the networks work any more. Content and networks are becoming 'virtualised' and 'replicated' across this master network itself, bringing the network itself closer to the end user. We don't need to traverse the whole internet for a query, because the internet is stored closer and closer to us. So, the image of the 'tree' is not quite right either. The branching network created by a CGNAT, i.e., my public IP address that is also shared by three of my neighbours, is not really a network in a meaningful sense. It is more of a technical 'workaround' for a shortage in IPv4 addresses. The important thing is that it _works_ because any performance costs it may have introduced have been massively negated by the other techniques introduced by CDNs. It doesn't matter that it doesn't fit with the old 'flat' architecture, because that old flat architecture is less and less operative at the macro level also. Whereas in the previous model, each node would have its own 'view' of the internet via its BGP/routes, now each _end user_ has their own 'view' of the internet through their personalised caches. So, we can't really say that the Internet is becoming _deeper_ (in the sense, say, that to get to a place you have to travel up branches and down other branches), but rather that it is becoming _multiple_ - i.e., its architecture remains relatively 'flat' (after all, NAT is not really translating from one _type_ of network to another - the network is still those 'dumb' pipes on each side), but is becoming more _fractal_[3]. =>[3] https://spool-five.com/box3/20220615t000005--the-hermeneutic-fractal__box3/