4 ms·

Gemini feels like a job half done: the presentation layer is simplified, but the TCP/TLS stack is still there in all its monstrosity, and so is the model that t

Gemini feels like a job half done: the presentation layer is simplified, but the TCP/TLS stack is still there in all its monstrosity, and so is the model that the user agent downloads stuff from the authoritative source live in response to interaction, with all the associated issues, like leaking meta-data. It also lacks a discoverability protocol improving on the search engine model. These are hard problems arguably.

selfhoster11 5y ago
TCP is not hard. It's not perfect, but for what Gemini needs (ordered guaranteed delivery of packets) it's a good tradeoff. Many firewalls would happily drop non TCP/UDP traffic as well, which would hurt adoption. For TLS, I agree it's a monstrosity, but it was an intentional protocol tradeoff. We really want to have privacy, so the only other way is to homebrew your crypto protocols (which is just as dangerous as homebrewing your crypto algorithms). Since we're basically stuck with TLS, Gemini asks the question of how to make TLS make a run for its money - and this is by using TLS asymmetric crypto keys for persistent or semi-persistant identity handled by the client. The HTTP web still doesn't have single sign-on, but Gemini does. Just present the same key everywhere. As for search, it's not like the HTTP web has a good discoverability story either. The only reason why we have search that can pull up any page based on any keyword is because we've spent decades selling our eyeballs to a company that indexed it all by sheer brute force. The Gemini web is arguably no different here - if someone wants to build a search engine, they are free to do so.
- baq 5y ago
  > Since we're basically stuck with TLS, Gemini asks the question of how to make TLS make a run for its money - and this is by using TLS asymmetric crypto keys for persistent or semi-persistant identity handled by the client. The HTTP web still doesn't have single sign-on, but Gemini does. Just present the same key everywhere. and the selling point is 'no tracking'? i must be missing something.
  - jlundberg 5y ago
    Well, the client don’t need to send the key all the time. Only when you want auth. And could easily implement one-keypair-per-server based on user preference.
- meltedcapacitor 5y ago
  Gemini needs neither ordered nor guaranteed delivery of packets, that's the point. It's a pretty context free format: each line can be rendered separately, and the user agent can display things as they arrive (with some visual indication of "holes" for lost packets while they're being re-requested). A partial page may also be enough for the user to decide they want to move on, saving pointless delivery of the entire page. Besides many Gemini pages fit in 1 or 2 packets, so there is more wastage through the TCP setup dance.
toast0 5y ago
TLS is kind of icky (so many size fields), but it's not really a monster if you limit your scope; you can use ciphers from elsewhere, maybe limit to TLS 1.3 if daring, or just 1.3 and 1.2 if you want broad (but not infinite) compatability. The monster is x.509 certificates.

selfhoster11 5y ago

TCP is not hard. It's not perfect, but for what Gemini needs (ordered guaranteed delivery of packets) it's a good tradeoff. Many firewalls would happily drop non TCP/UDP traffic as well, which would hurt adoption. For TLS, I agree it's a monstrosity, but it was an intentional protocol tradeoff. We really want to have privacy, so the only other way is to homebrew your crypto protocols (which is just as dangerous as homebrewing your crypto algorithms). Since we're basically stuck with TLS, Gemini asks the question of how to make TLS make a run for its money - and this is by using TLS asymmetric crypto keys for persistent or semi-persistant identity handled by the client. The HTTP web still doesn't have single sign-on, but Gemini does. Just present the same key everywhere. As for search, it's not like the HTTP web has a good discoverability story either. The only reason why we have search that can pull up any page based on any keyword is because we've spent decades selling our eyeballs to a company that indexed it all by sheer brute force. The Gemini web is arguably no different here - if someone wants to build a search engine, they are free to do so.

baq 5y ago
> Since we're basically stuck with TLS, Gemini asks the question of how to make TLS make a run for its money - and this is by using TLS asymmetric crypto keys for persistent or semi-persistant identity handled by the client. The HTTP web still doesn't have single sign-on, but Gemini does. Just present the same key everywhere. and the selling point is 'no tracking'? i must be missing something.
- jlundberg 5y ago
  Well, the client don’t need to send the key all the time. Only when you want auth. And could easily implement one-keypair-per-server based on user preference.
meltedcapacitor 5y ago
Gemini needs neither ordered nor guaranteed delivery of packets, that's the point. It's a pretty context free format: each line can be rendered separately, and the user agent can display things as they arrive (with some visual indication of "holes" for lost packets while they're being re-requested). A partial page may also be enough for the user to decide they want to move on, saving pointless delivery of the entire page. Besides many Gemini pages fit in 1 or 2 packets, so there is more wastage through the TCP setup dance.

toast0 5y ago

TLS is kind of icky (so many size fields), but it's not really a monster if you limit your scope; you can use ciphers from elsewhere, maybe limit to TLS 1.3 if daring, or just 1.3 and 1.2 if you want broad (but not infinite) compatability. The monster is x.509 certificates.