[pas]

> You don't see the same bs in other areas of computer science.

It's not like databases haven't been reinvented every time there's a new fancy startup.

MySQL the OG "but web scale" DB (even with InnoDB because consistency level set to - you guessed it - "web scale"), then all the NoSQL kids because of course "web scale", then SSD/NAND write-amplification and LSM and LevelDB and whatnot and nowadays RocksDB. And there's the memcached/Redis/Valkey.

It's not like networking is not doing the same. HTTP2, HTTP3. DNS over HTTPS, how amazing it is to establish a TLS connection just to send a base64 encoded DNS question. WTF. And there's a myriad things in eBPF land, all the kubernetes-related CNI providers, and during the OpenStack years we had new overlay networks every year (VXLAN, Geneve, and I forgot the name of the other stuff), and if you go a bit closer to datacenter networking and peek into the firehose of 802.1 of new protocols you'll drown from the constant buzz over which variation of the same thing is the obviously best one! (I still remember when people started to realize that maybe there's a better and cheaper way to avoid VLAN and spanning tree problems than "just buy Cisco forever and don't forget to enable rapid STP (802.1w)" ... and then for a while there was MSTP (802.1s), TRILL, and SPB (802.1aq) and now it's managed networks all the way down, I guess)

[jeltz]

While I agree with you in general I have to disagree about HTTP3. The reason we got it is because of middleboxes preventing the adoption of SCTP. And SCTP and QUIC solve very real needs in a good way, things which we could not solve with TCP.

Maybe Google instead should have used their politucsl clout to push for Http over SCTP. But it is not like there was a suitable predecessor to HTTP3.

[pas]

One, Zero, NextJS and so on all solve real problems on the level similar to what SCTP/QUIC does. That's my point. The seemingly endless churn has very good reasons. But of course the cost of the churn is still real too.