4 ms·

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

betaby 4y ago
Nothing to do with with operators. Again, physics, notably tcam https://en.wikipedia.org/wiki/Content-addressable_memory#Ternary_CAMs https://en.wikipedia.org/wiki/Content-addressable_memory#Ter... very expensive.
wmf 4y ago
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.
msbarnett 4y ago
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.
- netr0ute 4y ago
  Who said the prefixes had to go to /32? Just to /28 would only need 16 million routes.
  - lostapathy 4y ago
    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.
    - toast0 4y ago
      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.
      - lostapathy 4y ago
        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.
  - ikiris 4y ago
    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.
- lillecarl 4y ago
  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.
jleahy 4y ago
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).
- fomine3 4y ago
  It seems that Cisco distinguishes TCAM and Hash at least. https://quickview.cloudapps.cisco.com/quickview/bug/CSCvg60292 https://quickview.cloudapps.cisco.com/quickview/bug/CSCvg602... https://www.cisco.com/c/en/us/support/docs/switches/catalyst-9500-series-switches/217703-understand-hardware-resources-on-catalys.html https://www.cisco.com/c/en/us/support/docs/switches/catalyst...
  - jleahy 4y ago
    The documentation there is wrong/misleading.
zokier 4y ago
Getting with the times these days means adopting ipv6, not putting more bandaids to ipv4

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

wmf 4y ago

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.

msbarnett 4y ago

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.

netr0ute 4y ago
Who said the prefixes had to go to /32? Just to /28 would only need 16 million routes.
- lostapathy 4y ago
  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.
  - toast0 4y ago
    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.
    - lostapathy 4y ago
      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.
- ikiris 4y ago
  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.
lillecarl 4y ago
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.

jleahy 4y ago

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).

fomine3 4y ago
It seems that Cisco distinguishes TCAM and Hash at least. https://quickview.cloudapps.cisco.com/quickview/bug/CSCvg60292 https://quickview.cloudapps.cisco.com/quickview/bug/CSCvg602... https://www.cisco.com/c/en/us/support/docs/switches/catalyst-9500-series-switches/217703-understand-hardware-resources-on-catalys.html https://www.cisco.com/c/en/us/support/docs/switches/catalyst...
- jleahy 4y ago
  The documentation there is wrong/misleading.

zokier 4y ago

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