231–240 of 322 posts
Does this include laws about the internet? Because I think if the Computer Fraud and Abuse Act were altered somehow sufficiently that aaronsw hadn't been arrested, I'd prefer that future.
- Get rid of ARP - just append the LAN address to the network address like other networks. By default LAN addresses are random. (Note IPv6 enables this basically.)
- Support encrypted DNS and authenticated BGP.
- Let DNS return other metadata including the port as well as the IP address.
- Let DNS caching work. Don't misuse short DNS timeouts for load balancing.
- Ingress traffic filtering - reject source IP addresses from outside the current prefix.
- Not IP per se, but let multipath work in the LAN (and give Ethernet a TTL so that packets don't loop forever if things go bad.)
- Eliminate (or minimize) broadcasts. Use unicast/multicast for DHCP, service lookup, etc..
- Support relocation/forwarding of TCP connections so they don't break when your IP address changes.
- Fix TCP congestion control so that the data rate doesn't decrease as latency increases.
- Second adding congestion notification to TCP to differentiate between packet loss and congestion.
- Encrypt the host name in SSL/TLS.
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. - Y…
As an alternative:
- Private and federated. Everyone has a personal server application which spans multiple personal computing, storage, and peripheral devices and supports federated access at varying levels of security.
- The server stores and manages a user's private data and anything else they feel like storing or sharing. (This requires unobtanium level security, but since we're imagining let's pretend this is a solved problem and see where it takes us.)
- Sharing of all kinds is user controlled and is opt-in across multiple competing federated networks. This includes social networks - with the difference that anyone can start their own network, for any purpose.
- Networks are decentralised and peer-to-peer, and do not store personal data, track, or profile users. This is a user=centric network where users own and control their data. Not an industrial data silo network.
- Users can share different interest profiles and personal details across different networks with varying levels of security and implied credibility.
- Ads are opt=in not opt-out, and defined by voluntary and informed profile and interest sharing, not involuntary and uninformed data harvesting.
- Anonymous microtransactions are a thing. Anyone can sell at scale with as little friction as possible.
- There are no cryptocurrencies and no blockchain tech, because generating random numbers with the equivalent of your own electrical substation is fucking stupid. There is a low-energy secure equivalent. (See unobtanium. Or is it?)
- A common kit of essential server apps is open sourced and community-maintained.
- Commercial and/or professional apps are available by hire or subscription. Servers have a multi-profile multi-layer security model which controls which layer of personal and/or server data outsider apps have access to.
- All paid-for apps supply full details of the schemas and file formats they use, to guarantee that users can freely transfer data to a competing app provider so apps and services hold personal data hostage and have to compete on service quality, not on retention gaming.
- Hacking, malware, virus creation, phishing, and so on, are punished by deletion of personal server data and reduction to the most basic server hardware and software. For serious and repeat offenders, this is for life.
- IoT devices are treated as personal server peripherals with no external data sharing (except by opt-in.)
- Government and military networks use an expanded version of the same system. Municipal, military, and internal gov services run on separate private subnetworks which can only be accessed through authorised devices with extra ID verification, not through general public logins.
Basically it's a combination of device security, private ID (probably biometric), sacrosanct personal data protection, high user-controlled privacy, super low cost of entry for entrepreneurial service provision, squashing of local, national and international scales, and strong forcing of anti-monopolistic competition - the opposite of the current model, which seems to be about herding users into virtual pens owned by monopolists, applying various psychological patterns to control and trigger behaviour, monitoring behaviour and sentiment through minimal privacy, and having to deal with very leaky and insecure devices and systems.
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. - Y…
> 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.
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."
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. - Y…
> Mail is not forwarded Go one further, mail is pulled, not sent. One such approach is 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.
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. - Y…
> 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.
I would fix the order of URLs instead of http://news.ycombinator.com/item it would be https:com/ycombninator/news/item
com.ycombinator.news/item
On the other hand, you could do an incremental query perhaps and cache the path length.
But I wonder about having the protocol name in there. Couldn't DNS return the ports for the supported services and then you could pick whichever one you want?
Honestly I think it couldn’t have gone a lot better... The worst that happened to the internet is Google becoming evil. The internet circa 2000 was mind blowing.
It was mind blowing indeed. It was such an adventure building stuff on the web back then. We didn't have the choice we have today. I don't think Google became "evil". It just realised (or remembered) it's a business.
Exactly, it became evil!
So so many protocols (history), but how few could we get away with, and what would they look like?*
Why don't we have constructive computational contracts for computational work?
How do we make the Internet easier to understand?
How do we manage the agency problem? We yield far too much agency as a matter of daily life, our data is not our own, our decisions are shared with barely knowable third parties.
How do we design human computer interfaces with health, especially mental health, as primary constraint?
How rapidly can the EU coalesce around a combination of a RISC-V general purpose CPU (with suitable trimmings) and a SEL4-influenced-kernel, perhaps in Rust (https://gitlab.com/robigalia)?
How do we standardise on constraints of discourse such as those pertaining to offensive language or hate speech? How do we make it easier for people to communicate with kindness? Autohinting everywhere? Like a shellcheck for human bashfulness?
VR and AR are coming very soon and without care they will be shatteringly destructive of human life. Humans addicted to computationally modeled utility functions mediated by multi-sensory computer games?
How do we embed the lore in the experience? How do we make available all the references as delightful marginalia?
What areas of Mathematics and Physics do we need to study to get ahead of our problems? Category theory is beyond trendy, what's trending? How about rigorous dimensional analysis to match the type theory, or sumthin? How do we invite the world's smartest financiers to apply and share their thought more generously?
Can we settle on a basic curriculum? What functional minimum of linguistic, mathematical, physical, visual, and other skills do we need? Is lisp or a variant the first language we should learn, and if so how should we be able to learn it? If not lisp then what? APL? Fortran? Compiler forbid, Haskell?
How do we ensure that code and documentation are always in sync? How much time will this require?
How do we guarantee a standard of professional attainment and delivery of ICT expert that is globally effective? How do we standardise how we do, not just what we do?
How best can we help each other make our Internet an even better place?
(much spelling, apologies)
*4