> I don’t see an easy way for a hedge fund to rent out the IP space unless they are a cloud provider.

This is a solved problem these days.

I personally know of IPXO[0] and can vouch for them as being legitimate and well ran. If you have an idle /20 laying around or whatever, I highly suggest getting it on their market and making some relatively easy money leasing out your blocks.

The total interaction (short of initial setup) these days is clicking a few buttons in your favorite interface to publish ROA records every so often when an IP block sees churn. A hedge fund could hire a junior neteng to handle those requests, the very rare abuse complaint, and any hijacked routes you need to track down.

Overall this is becoming a mature market with a number of players emerging for a public market. In private it's been going on for some time on a much more informal basis.

[0] ipxo.com

Actually, you can assign (not transfer) IP ranges and the one advertising it does not need to own it, but would need a Letter of Authorization (LoA) or similar.

What causes the restriction of /24, rather than being able to route individual addresses?

Physics.

Routers can only handle so many routes, so their operators set floor values for size of route they'll generally accept.

Then it's time for operators to get with the times, because routers today don't have those same limitations as before.

Recent routers can handle 1M routes while the Internet routing table currently has... 928K routes. Allowing people to disaggregate further would blow out routers.

I'm not counting gold-plated routers because I see no reason to force ISPs to buy those.

Nothing to do with with operators. Again, physics, notably tcam https://en.wikipedia.org/wiki/Content-addressable_memory#Ter... very expensive.

TCAMs as shown in that page haven’t been used for a very very long time. Even in basic switches now the solution is ‘algorithmic TCAM’ (ie. something like hashtables or tries).

Huge numbers of routes are simply not an issue these days, unless you have truly ancient crappy equipment (maybe it was second hand for example).

It seems that Cisco distinguishes TCAM and Hash at least. https://quickview.cloudapps.cisco.com/quickview/bug/CSCvg602... https://www.cisco.com/c/en/us/support/docs/switches/catalyst...

The documentation there is wrong/misleading.

Do you have any conception of how much TCAM would be required to individually route 14 billion IP addresses? And even beyond that what else would be involved to route things fast enough at that granularity.

A top end router today can handle ~million routes. You’re talking a four orders of magnitude increase.

I'm Linux there's a thing called route cache, which is really fast. Because in reality, even if you're a router at an IX you won't see traffic to all billions of ips.

So if an ip isn't in the cache it'll go through a slow path instead and get cached.

Why this wouldn't be possible in hardware is beyond me.

Routing on commodity hardware is becoming a thing too with DPDK and OVS, you can now do 200gbps on a single x86 box.

Who said the prefixes had to go to /32? Just to /28 would only need 16 million routes.

I'm sure if you start a fund to pay those operators to care, they'll get right on it. The current case is the people who need much more granular space are different than the people who'd bear the cost of it.

Otherwise, just move to IPv6 or something and fix the problem in a saner way.

Practical reasons of replacing "all the routers" aside ... would this even be desirable?

Do we really want IP space carved up into smaller and smaller bites? Do we really want even more and smaller entities advertising routes into BGP? The big players have enough troubles with it - we don't need small businesses with a /28 playing in those waters.

Kind of / maybe. At my last job, we wanted to host our own anycast nameservers. You get the best results from having four nameservers in DNS, and you want them independent, so there's four /24s out there doing not much other than DNS (I think there may be some other stuff now). On the other hand, some other changes meant our thousands of servers no longer had their own ipv4 addresses, so there's several /24s returned to the hosting provider to be used by others.

There's a lot of value in owning your user-facing IPs and being able to move them around as needed, but a lot of services don't really have that many user facing IPs, so utilization is low if they get a /24 per location.

I tend to think the amount of companies who want to do that AND could do it well is vanishingly low.

On the converse - I think of all the smaller businesses I worked around that really wanted to get their own portable IP block and ASN because they thought it made them serious ... but just had no business playing in that space.

Getting with the times these days means adopting ipv6, not putting more bandaids to ipv4

Routing tables need to be limited in size in order for hardware to decide where to route packets fast enough. In the past hardware was more cpu and memory bound than the present.

But also at present, dealing with millions of routes at the packet-switching level gets difficult. Especially because the speed of links has increased so much.