3 ms·

I am confused. The major problem is claimed to be un/marshalling. But they take no closer looks. Advise an architecture without it just to introduce another wit

by vast 7y ago

I am confused. The major problem is claimed to be un/marshalling. But they take no closer looks. Advise an architecture without it just to introduce another with it. Seems like protocol overhead will stay.

shereadsthenews 7y ago
Yeah, the only thing we can really learn from this paper is that the authors aren't very familiar with protobuf. Anybody who claims it takes 10 microseconds to decode a 1KiB protocol message is either intentionally constructing the worst-case message or doesn't know what they are talking about.
- justin66 7y ago
  > Yeah, the only thing we can really learn from this paper is that the authors aren't very familiar with protobuf. This is stupid. Three of the four authors are engineers at Google.
  - hinkley 7y ago
    That should not surprise you. When Microsoft was at its dumbest, its corporate identity was that they were hiring the best of the best. Now Google has that baton. Wonder who they’ll pass it to...
    - justin66 7y ago
      I did not intend to express agreement with the OP's stupid comment.
sitkack 7y ago
If marshalling is the problem, fix marshalling. Erlang (built in distribution), CapnProto, Flatbuffers, etc https://capnproto.org/news/2014-06-17-capnproto-flatbuffers-sbe.html https://capnproto.org/news/2014-06-17-capnproto-flatbuffers-...