Unlocking QUIC’s proxying potential with MASQUE(blog.cloudflare.com)
blog.cloudflare.com
Unlocking QUIC’s proxying potential with MASQUE
https://blog.cloudflare.com/unlocking-quic-proxying-potential/
3 comments
QUIC datagrams not having a stream ID was a compromise, which is why the H3-DGRAM draft exists to add them. Any other protocol can use cite and use H3-DGRAM even if it itself is not using HTTP/3.
what possible worse position could there have been? i dont understand why anyone would protest basic common sense of having a multiplexing protocol do multiplexing.
making it a hybrid multiplexing/not-multiplexong protocol, that can easily wrap & encapsulate streams but which is dead ass useless & requires subprotocol negotiation & tracking to deal with anything else seems like unbelievable busted jank. i literally cant imagine a worse way to do this. how was this compromise? who does this satisfy? i would have told any extremist protesting common semse to take a hike, would have started my own draft, would have died on this hill.
this is murderously bad incidental complexity. it's uncontainedly bad. what if someone doesnt opt to use the h3-dgram variant? then we are up shits creek: our protocols conflict & there's no way to resolve this mess. how do we differentiate channels (my term for "streams" of unordered datagrams) if everyone has different ways of defining channels? we invite in infinite forms of incompatibility. maybe we need to revamp the h3 encapsulation & spinkle a couple magic bytes as preamble, to try to mark our turf, declare our subprotocol.
this is the definition of insanity. this is a stupid awful mess. irreconcilable & infinitely confusing, forever & ever. disastrously wrong. how possibly could datagrams be done worse? what was "compromised"?
i have had such high hopes QUIC would be the transport protocol to encapsulate everything, to make a new dawn of connectivity possible. this is weap-into-my-keyboard stupid/bad news, this is impossibly broken, this is disaster: DATAFRAME is broken by design, due to "compromise", and nothing non-stream-based can work together ever atop QUIC (except in some special lucky cases) because the basic, most core, simple, sensible-est of guiding principles of QUIC got thrown out the door to make DATAFRAME, and it's ruin & damnation for it. DATAFRAME could not have a more elementary mistake if it tried. sctp notably managed to avoid shooting themselves in a vital area like this. abandon this DATAFRAME effort. it jeopardizes the overall QUIC effort itself that it is so awful.
making it a hybrid multiplexing/not-multiplexong protocol, that can easily wrap & encapsulate streams but which is dead ass useless & requires subprotocol negotiation & tracking to deal with anything else seems like unbelievable busted jank. i literally cant imagine a worse way to do this. how was this compromise? who does this satisfy? i would have told any extremist protesting common semse to take a hike, would have started my own draft, would have died on this hill.
this is murderously bad incidental complexity. it's uncontainedly bad. what if someone doesnt opt to use the h3-dgram variant? then we are up shits creek: our protocols conflict & there's no way to resolve this mess. how do we differentiate channels (my term for "streams" of unordered datagrams) if everyone has different ways of defining channels? we invite in infinite forms of incompatibility. maybe we need to revamp the h3 encapsulation & spinkle a couple magic bytes as preamble, to try to mark our turf, declare our subprotocol.
this is the definition of insanity. this is a stupid awful mess. irreconcilable & infinitely confusing, forever & ever. disastrously wrong. how possibly could datagrams be done worse? what was "compromised"?
i have had such high hopes QUIC would be the transport protocol to encapsulate everything, to make a new dawn of connectivity possible. this is weap-into-my-keyboard stupid/bad news, this is impossibly broken, this is disaster: DATAFRAME is broken by design, due to "compromise", and nothing non-stream-based can work together ever atop QUIC (except in some special lucky cases) because the basic, most core, simple, sensible-est of guiding principles of QUIC got thrown out the door to make DATAFRAME, and it's ruin & damnation for it. DATAFRAME could not have a more elementary mistake if it tried. sctp notably managed to avoid shooting themselves in a vital area like this. abandon this DATAFRAME effort. it jeopardizes the overall QUIC effort itself that it is so awful.
> DATAGRAM frames do not contain a multiplexing identifier, unlike QUIC's stream IDs.
This sounds like the stupidest plan ever. There's a nice multiplexing protocol, but as soon as you want unreliable or datagram messages, you need to re-invent a new application level protocol for routing datagrams to their specific consumers? This is the dumbest shit.
There should certainly be different channels of datagrams flowing in quic. How we would think for a second anything else makes sense is is uncomprehensible. What the frak?
And indeed, MASQUE right away hits this inconcievably stupid transport limit & has to invent their own "quarter stream id" to figure out where given datagrams are coming from or going. Anytime a unreliable protocol needs to stream over quic, it too is going to have to invent it's own similar transport protocol wrapper, rather than be able to simply just work directly atop quic. This is such a vastly stupid thing, criminal negiglence in protocol design, is a vast world of infinite suck. Quic datagrams, you screwed up! Wicked bad!