Jow Forums reinvents TCP/IP

Alright, fa/g/s, how would you rework TCP/IP?

Attached: 06E20F43-65F4-4E5B-A974-13164FA15E44.jpg (1003x624, 107K)

Other urls found in this thread:

apenwarr.ca/log/20170810
quicwg.org/base-drafts/draft-ietf-quic-http.html
youtube.com/playlist?list=PLowKtXNTBypH19whXTVoG3oKSuOcw_XeW
twitter.com/SFWRedditVideos

Add a blockchain layer.

I wouldn't. It's fine.

don't fix what isn't broken

is it weird to say that i have a problem with understanding tcp/ip .

i read about it , but cant put it in practice .
i read a lot , but am feeling like i dont know what to do with it or how to do SOMETHING WITH IT ??

can someone relate ?

Attached: cat.jpg (641x581, 56K)

Add evil bit

u prolly lack more fundamental knowledge

>:3

I’m sure the great minds at 4channel can find some areas for improvement

i would flip it upside down

what is a good book to learn tcp/ip?
and the linux networking stack?

What exactly do you want to do with it, my negro friend

Handshake is now 2-way.

I'd just remove it and force applications to bring their own reliable transport implementation.

TCP was a fucking mistake.

what if we made the physical port like a lot smaller

Any good audiobooks or talks for TCP/IP? I’m too ADHD to read a 2,600 page book

Attached: E600FE67-48AA-4B18-8F9A-F4FB37105DBA.jpg (3088x2320, 1.25M)

brainlets.

For one add a security layer.

t. Doesn't understand IP

Oh IP is so fucking bad. Legacy bloat the protocol. apenwarr.ca/log/20170810

Encryption in core, there is a lot of mess from control packets not being authenticated.
The model would be that client is required to know server's longterm pubkey and first packet is already authenticated. CurveCP and Noise protocol are a nice examples.
Unlike TCP, I would love it to be save for tunneling it over itself.
Potentially to have a mode that allows packet loss. There is shit ton of stateful unreliable protocols over UDP that just want this, hopefully such optional mode would provide them a unified ground. Single semi-stateful protocol would have an advantage of high adoption and would thus mostly work with firewalls.

Attached: Screenshot from 2019-08-15 18-53-04.png (868x218, 51K)

I also have no idea what QUIC brings in and what are the advantages, maybe I could learn something from it. Well one more TODO thing on a list.

2^32 ports for the ultimate CGNAT

QUIC is over UDP. It is reliable. It is designed to have less overhead and lower latency than TCP. It also handles unstable connections better.

I'd use elven magic by enslaving elves.
The magic is most powerful when they orgasm.

Attached: 1553040280152.jpg (499x393, 90K)

Change the data link layer from MAC to PCMASTERRACE

that's a very incomplete description to the point of uselessness. why did you even bother?

>that's a very incomplete description to the point of uselessness. why did you even bother?
user's post:
>I also have no idea what QUIC brings in and what are the advantages
Looks like I answered user's question as it was written.

Pretty much the whole underlying network protocols of the internet are retarded, not fit for purpose and made by overly invested nerds in the 1970s who did shit in stupid ways because they had penis size issues.
Throw it all out. Throw out TCP/IP, HTTP, HTTPS, DNS, WWW, UDP, everything. It's all fucking crap

and user found the answer useless because it was very incomplete description.

these were designed for 56.6kbps internet at most

the overhead is non issue nowadays with comparably absurd internet speeds, if it's working why fix it?

quicwg.org/base-drafts/draft-ietf-quic-http.html

Add path-like strings instead of ports.

t. massive brainlet

Attached: 1509200876056.jpg (645x729, 49K)

How do you know that was the same user?

because samefag on all three posts

just replace it with SCTP

multicast all the things, no addressing needed. Everything layer 3

Great idea, remove the port and add in internet via Bluetooth.

Attached: e0a.gif (2952x3047, 1.18M)

Remove everything between application layer and physical layer. It's all bloat anyways

TCP is a perfect __generic__ end-to-end protocol for reliable delivery. Of course some applications may benefit from changes, but as a generic protocol, TCP is king.

Look up the Lindy effect, you faggots.

youtube.com/playlist?list=PLowKtXNTBypH19whXTVoG3oKSuOcw_XeW

Based quicposter.

okay, but what is a good book to learn tcp/ip?
and the linux networking stack?

jesus christ, broke me on the first reply

The government approval verification layer.

I make the logo

>another cable guy thread
Just face it, you are among the first that get replaced by AI, even now WiFi is pretty much the standard, not to mention when 5g finally takes off.

tcp/ip get a cisco cert

>how would you rework TCP/IP
Make it 7 layers by splitting the "Application" layer into Application, Presentation and Session layers.

Merge L2 and L3, IPv6 is already making similar changes to that with ICMPv6 might as well remove L2 completely. Bake encryption and authentication into IP.

how can we use docker on it?

>Make IPv7. It has 64 bit addresses and the entire existing IPv4 space is routable as 00.00.00.00.XX.XX.XX.XX.
Fuck IPv6, what the hell were they thinking?
>Replace TCP with SCTP.
>Make UDP only checksum the packet header, not the body.
>Redesign everything to support 16kB packets without fragmentation.
That's what I can think of off the top of my head.