5 ms·
Changes I would have made in the early days: - 48-bit static IP addresses. 70 trillion should be enough. 128 bits was overkill. - Nodes, not interfaces, have
by Animats 5y ago
Changes I would have made in the early days:
- 48-bit static IP addresses. 70 trillion should be enough. 128 bits was overkill.
- Nodes, not interfaces, have IP addresses, so you can use multiple paths.
- IPSEC available but initially optional.
- Explicit congestion notification, so packet loss and congestion loss can be distinguished.
- Everything on the wire is little-endian, byte oriented, and twos complement.
- You can validate a source IP address by pinging it with a random number. If you don't get a valid reply, the IP address is fake. Routers do this the first time they hear from a new address, as a form of egress filtering. This contains DDOS attacks.
- Routers will accept a "shut up" request. If A wants to block B, it sends to a router on the path, the router pings A to validate the source, and then blocks traffic from B to A for a few minutes. This also contains DDOS attacks. Routers can forward "shut up" requests to the next router in the path, for further containment.
- Fair queuing at choke points where bandwidth out is much less than bandwidth in.
- Explicit quality of service. At a higher quality of service, your packets get through faster, but you can't send as many per unit time.
- No delayed ACKs in TCP.
- Fast connection reuse in TCP.
- Mail is not forwarded. Mail is done with an end to end connection. Mail to offline nodes may be resent later, but the sender handles that. Mail, instant messaging, and notifications are the same thing. Spam is still possible but hard to anonymize. If you want your mail buffered, use an IMAP server at the receive end.
- One to many messaging uses a combination of RSS and notifications.
- Something like Gopher should be available early. The Web would not have fit in early machines. but Gopher would.
- pteraspidomorph 5y ago> Nodes, not interfaces, have IP addresses, so you can use multiple paths. Isn't this difference academic/software? What is the crucial point here that can no longer be implemented as an abstraction? (Not a networking specialist so this might just be my ignorance speaking.) > You can validate a source IP address by pinging it with a random number. Good idea, but I suspect this is the type of feature that might quickly fall prey to implementation laziness/incompetence, or in other words, enough implementers would both ignore it and not use it until it became unreliable. > Routers will accept a "shut up" request I like this idea but couldn't this system be abused by a malicious man in the middle to much more easily hinder connectivity to a targeted system? The man in the middle can fake the validation too... > Mail is not forwarded I don't think this is a good idea. People would quickly invent a forwarding protocol if one didn't exist. I mean, I see what you're trying to do here, but I don't think you would succeed on this one.
- Buttons840 5y ago>> Routers will accept a "shut up" request > I like this idea but couldn't this system be abused by a malicious man in the middle to much more easily hinder connectivity to a targeted system? The man in the middle can fake the validation too... I think of it more as a "I'm just ignoring these, you mine as well stop sending them" request. After all, a man-in-the-middle can already just drop the packets.
- benlivengood 5y ago> I think of it more as a "I'm just ignoring these, you mine as well stop sending them" request. After all, a man-in-the-middle can already just drop the packets. Some MITM can only inspect and forge traffic, not drop it (e.g. NSA tapping fiber)
- twobitshifter 5y agoRe: mail - I believe what gp is saying is that mail will be sent directly - made possible by the larger IP address space. Once it’s on your device you can copy and move it wherever. This would actually match the original design of email before it became the case where everything was via pop and imap. Of course the biggest flaw with email was not expecting spam. My change would be accepting messages only from known senders or those who have done the needed “work” https://www.w3.org/2003/10/acquaintance-protocol/ https://www.w3.org/2003/10/acquaintance-protocol/
- darkr 5y ago> 48-bit static IP addresses. 70 trillion should be enough. 128 bits was overkill The least significant 64 bits in IPv6 are used for link-layer addressing, which is how IPv6 supplants older L2 protocols (ARP etc) with NDP etc, thereby fixing (or at least improving) some significant issues with larger IPv4 subnet scalability/reliability
- chalst 5y agoRight. The reference IPv6 packet header is only twice as long as that for IPv4 (320 bits vs 160). What is crucial, combined with the fact darkr points out that since they include the link layer the IPv6 protocol deals with more of the network architecture, is that IPv6 headers are much simpler, with only 8 fields, the 6 non-address fields all in the first 64 bits, while IPv4 has 14 fields, with the 12 non-address fields taking 96 bits, and they are better designed with attention to what routers find most important. This simplicity offers routing subsystems increased flexibility in the way they can be put together and handle their workloads.
- kodablah 5y ago> Mail is not forwarded Go one further, mail is pulled, not sent. One such approach is https://tonsky.me/blog/streams/ https://tonsky.me/blog/streams/. It'd be hard to prevent moats and people might still coalesce into a few popular middlemen, but at least spam would be reduced. Spam is the primary reason client-side/self-hosted mail is left to the tech-savvy.
- mjevans 5y agoI prefer the "headers are sent" but "messages are pulled" variation. Like downloading from NNTP.
- throwaway81523 5y agohttp://cr.yp.to/im2000.html http://cr.yp.to/im2000.html
- FranchuFranchu 5y agoI'd include some extra bits in IP addresses to eliminate NAT
- science4sail 5y ago> 48-bit static IP addresses. 70 trillion should be enough. 128 bits was overkill. What if humanity becomes a Kardashev type-2 or an interstellar civilization? In that case the population of humans, much less computers, could easily succeed 70 trillion.
- tgbugs 5y agoIt is unlikely that we will be able to use IP as a way to create a single network between Earth and Mars much less outside the solar system.
- hansel_der 5y agonot unlikely that we will have ip-islands connected by some DTN/sneakernet thou.
- solatic 5y ago> Routers will accept a "shut up" request This shifts the DoS threat from servers to clients. How does the router know that the request from A to block B is legit? Just because the router can ping A? Behold, a system where you can shut down your enemies for good by paying botnets to send requests to block B.
- mjevans 5y agoIt's implied that the block is ONLY for traffic TO A FROM B; via R (router) (with R having the chance to aggregate? Maybe based on owning delivery paths that are valid?). Full quote: "- Routers will accept a "shut up" request. If A wants to block B, it sends to a router on the path, the router pings A to validate the source, and then blocks traffic from B to A for a few minutes. This also contains DDOS attacks. Routers can forward "shut up" requests to the next router in the path, for further containment."
- solatic 5y agoThis is slightly disingenuous. If I'm a router manufacturer, and there's a protocol that will tell me "peers such-and-such are troublemakers", and I get twenty such notifications from downstream that peer Foo is being a troublemaker, then I can automatically block all traffic from peer Foo (forget about what the protocol actually intended) and actually sell that as a security feature.
- soneil 5y agoSo many conversations like this wouldn't have existed if the internet had been designed for zero-trust from day one. Imagine if all icmp / igmp had the possibility to sign requests.
- kazinator 5y ago> Nodes, not interfaces, have IP addresses, so you can use multiple paths. Operating systems would quickly develop "multi noding"; binding multiple node identities to a host, and then partitioning those among the interfaces. Separate IPs for different adapters is a good idea, and needed for NAT: e.g. a home router being 192.168.1.1 inward-facing, and having some external IP. I
- tptacek 5y agoThis is a great list. I don't think IPSEC would have been a win; it's not a great protocol just as a cryptographic transport, and I think one thing we've learned over the last 30 years is that the end-to-end argument applies especially well to cryptographic security (just because there are so many different service models you can want, and no one design that serves all of them).
- kobalsky 5y ago> 48-bit static IP addresses a static IP per device is google and facebook's wet dream. yeah apple, keep your IDFA for yourself. also, imagine getting banned and not being to use google anymore, at all. asking your kids to google something for you and getting them banned too.
- dublin 5y agoIn a great many cases, malware can grab your MAC address now. Which is, as it happens, a 48-bit globally unique address. IPv6 addresses are a major PITA, not only are they too long, they're impossible to write and get even most technically competent people to understand. They're almost as hideous and unusable as x.400 email addresses...
- 1vuio0pswjnm7 5y ago"- Mail is not forwarded." I experimented with this one over 10 years ago, pre-Wireguard. The idea was to first establish a peer-to-peer connection, e.g., via L2 overlay, then have each peer run their own smptd. A supernode runs on a publicly reachable server at a hosting company and is only necessary for extablishing a peer-to-peer connection, not for routing traffic. Each peer has an Ethernet interface, e.g., a tap device with a private address for the L2 overlay, and each peer runs their own smptd. As for spam, the trick is to limit the size of the overlay network. Users might belong to several L2 networks, e.g., work, home, school, etc. If it networks are kept small and the peers do not give out their email addresses to people not on the network, then it's possible to have a small, spam-free email network among the peers. If end-to-end email on small disparate networks became the norm, spammers would have to find all these small overlay networks and infiltrate every one. As I understand it, this was the original intended design of email: smtpd's directly communicating with each other. This still happens but users do not run smtpd's. Instead they were asked to run pop clients.
- caleb-allen 5y ago> Nodes, not interfaces, have IP addresses, so you can use multiple paths. SCTP has multi-path routing, no?
- ryeguy_24 5y agoTo be able to make these suggestions means you must have significant experience and knowledge of the internet backbone. Super impressive.
- _nub3 5y ago> Routers will accept a "shut up" request. If A wants to block B, it sends to a router on the path, the router pings A to validate the source, and then blocks traffic from B to A for a few minutes. This also contains DDOS attacks. Routers can forward "shut up" requests to the next router in the path, for further containment. This is an invitation to silence unwanted users by certain governments.
- albuic 5y agoYou might have missed the "from B to A" in the parent comment which is quite important. Moreover governments have other (legal) ways to silence users.